Cow behaviour pattern recognition using a three-dimensional accelerometer and support vector machines

Goats on the Move: Evaluating Machine Learning Models for Goat Activity Analysis Using Accelerometer Data

Putting sensors on the bodies of animals to automate animal activity recognition and gain insight into their behaviors can help improve their living conditions. Although previous hard-coded algorithms failed to classify complex time series obtained from accelerometer data, recent advances in deep learning have improved the task of animal activity recognition for the better. However, a comparative analysis of the generalizing capabilities of various models in combination with different input types has yet to be addressed. This study experimented with two techniques for transforming the segmented accelerometer data to make them more orientation-independent. The methods included calculating the magnitude of the three-axis accelerometer vector and calculating the Discrete Fourier Transform for both sets of three-axis data as the vector magnitude. Three different deep learning models were trained on this data: a Multilayer Perceptron, a Convolutional Neural Network, and an ensemble merging both called a hybrid Convolutional Neural Network. Besides mixed cross-validation, every model and input type combination was assessed on a goat-wise leave-one-out cross-validation set to evaluate its generalizing capability. Using orientation-independent data transformations gave promising results. A hybrid Convolutional Neural Network with L2-norm as the input combined the higher classification accuracy of a Convolutional Neural Network with the lower standard deviation of a Multilayer Perceptron. Most of the misclassifications occurred for behaviors that display similar accelerometer traces and minority classes, which could be improved in future work by assembling larger and more balanced datasets.

Classification of Behaviour in Conventional and Slow-Growing Strains of Broiler Chickens Using Tri-Axial Accelerometers

Behavioural states such as walking, sitting and standing are important in indicating welfare, including lameness in broiler chickens. However, manual behavioural observations of individuals are often limited by time constraints and small sample sizes. Three-dimensional accelerometers have the potential to collect information on animal behaviour. We applied a random forest algorithm to process accelerometer data from broiler chickens. Data from three broiler strains at a range of ages (from 25 to 49 days old) were used to train and test the algorithm, and unlike other studies, the algorithm was further tested on an unseen broiler strain. When tested on unseen birds from the three training broiler strains, the random forest model classified behaviours with very good accuracy (92%) and specificity (94%) and good sensitivity (88%) and precision (88%). With the new, unseen strain, the model classified behaviours with very good accuracy (94%), sensitivity (91%), specificity (96%) and precision (91%). We therefore successfully used a random forest model to automatically detect three broiler behaviours across four different strains and different ages using accelerometers. These findings demonstrated that accelerometers can be used to automatically record behaviours to supplement biomechanical and behavioural research and support in the reduction principle of the 3Rs.

Monitoring Cattle Ruminating Behavior Based on an Improved Keypoint Detection Model

Cattle rumination behavior is strongly correlated with its health. Current methods often rely on manual observation or wearable devices to monitor ruminating behavior. However, the manual monitoring of cattle rumination is labor-intensive, and wearable devices often harm animals. Therefore, this study proposes a non-contact method for monitoring cattle rumination behavior, utilizing an improved YOLOv8-pose keypoint detection algorithm combined with multi-condition threshold peak detection to automatically identify chewing counts. First, we tracked and recorded the cattle's rumination behavior to build a dataset. Next, we used the improved model to capture keypoint information on the cattle. By constructing the rumination motion curve from the keypoint information and applying multi-condition threshold peak detection, we counted the chewing instances. Finally, we designed a comprehensive cattle rumination detection framework to track various rumination indicators, including chewing counts, rumination duration, and chewing frequency. In keypoint detection, our modified YOLOv8-pose achieved a 96% mAP, an improvement of 2.8%, with precision and recall increasing by 4.5% and 4.2%, enabling the more accurate capture of keypoint information. For rumination analysis, we tested ten video clips and compared the results with actual data. The experimental results showed an average chewing count error of 5.6% and a standard error of 2.23%, verifying the feasibility and effectiveness of using keypoint detection technology to analyze cattle rumination behavior. These physiological indicators of rumination behavior allow for the quicker detection of abnormalities in cattle's rumination activities, helping managers make informed decisions. Ultimately, the proposed method not only accurately monitors cattle rumination behavior but also provides technical support for precision management in animal husbandry, promoting the development of modern livestock farming.

Determining energy expenditure in a large seabird using accelerometry

The trade off between energy gained and expended is the foundation of understanding how, why and when animals perform any activity. Based on the concept that animal movements have an energetic cost, accelerometry is increasingly being used to estimate energy expenditure. However, validation of accelerometry as an accurate proxy for field metabolic rate in free-ranging species is limited. In the present study, Australasian gannets (Morus serrator) from the Pope's Eye colony (38°16'42″S 144°41'48″E), south-eastern Australia, were equipped with GPS and tri-axial accelerometers and dosed with doubly labelled water (DLW) to measure energy expenditure during normal behaviour for 3-5 days. The correlation between daily energy expenditure from the DLW and vectorial dynamic body acceleration (VeDBA) was high for both a simple correlation and activity-specific approaches (R2=0.75 and 0.80, respectively). Varying degrees of success were observed for estimating at-sea metabolic rate from accelerometry when removing time on land using published energy expenditure constants (R2=0.02) or activity-specific approaches (R2=0.42). The predictive capacity of energy expenditure models for total and at-sea periods was improved by the addition of total distance travelled and proportion of the sampling period spent at sea during the night, respectively (R2=0.61-0.82). These results indicate that accelerometry can be used to estimate daily energy expenditure in free-ranging gannets and its accuracy may depend on the inclusion of movement parameters not detected by accelerometry.

Validation of inertial measurement units to detect and predict horse behaviour while stabled

BACKGROUND

Musculoskeletal injuries are observed in Thoroughbred racehorses and may become catastrophic. Currently, there are limited methods for early detection of such injuries. Most injuries develop gradually due to accumulated damage, providing the opportunity for early detection. Horses experiencing pain or lameness may exhibit changes in behaviour so the development of an objective, real-time system monitoring horse behaviour may enable detection of bone injuries before catastrophic failure.

OBJECTIVES

To determine whether intensive observational methods of assessing horse behaviour can be replaced by use of inertial measurement units (IMUs).

STUDY DESIGN

Validation study assessing IMU use against video observation.

METHODS

Six hospitalised Thoroughbreds (algorithm training data) and 19 Thoroughbred racehorses in-training (algorithm testing data) were equipped with an IMU placed on the lateral side of both forelimbs (left fore, LF; right fore, RF) and monitored in a stable for 4 h. An algorithm was developed to classify behaviour and then validated against video recordings.

RESULTS

Standing was the most prevalent behaviour (LF 88.8%, 95% confidence interval [CI] 88.7-89.0; RF 88.5%, 95% CI 88.4-88.7). IMU classification of recumbent and standing activities showed excellent agreement (sensitivity) with video observation (>98%). This was followed by stepping (LF 89.4%, RF 85.5%) then weight-shifting (LF 54.3%, RF 61.5%). Predictions from the algorithm showed misclassification of 2.5% (LF 5500/225 352, RF 5218/210 170). Excluding standing, misclassification was 6.8% (1705/25 158) and 7.5% (1812/24 077) for the left and right forelimbs, respectively, with pawing and weight-shifting most frequently misclassified.

MAIN LIMITATIONS

Increasing the number of horses and types of behaviours observed may improve predictions.

CONCLUSIONS

IMUs displayed a high sensitivity to movement on a small number of horses, and with further validation they have the potential to effectively monitor behaviour of racehorses in training. However, more sensitive methods may be needed to validate the classification of weight-shifting behaviour. Future studies should evaluate the association between each behaviour and musculoskeletal injury.

Precision Livestock Farming Research: A Global Scientometric Review

Precision livestock farming (PLF) utilises information technology to continuously monitor and manage livestock in real-time, which can improve individual animal health, welfare, productivity and the environmental impact of animal husbandry, contributing to the economic, social and environmental sustainability of livestock farming. PLF has emerged as a pivotal area of multidisciplinary interest. In order to clarify the knowledge evolution and hotspot replacement of PLF research, based on the relevant data from the Web of Science database from 1973 to 2023, this study analyzed the main characteristics, research cores and hot topics of PLF research via CiteSpace. The results point to a significant increase in studies on PLF, with countries having advanced livestock farming systems in Europe and America publishing frequently and collaborating closely across borders. Universities in various countries have been leading the research, with Daniel Berckmans serving as the academic leader. Research primarily focuses on animal science, veterinary science, computer science, agricultural engineering, and environmental science. Current research hotspots center around precision dairy and cattle technology, intelligent systems, and animal behavior, with deep learning, accelerometer, automatic milking systems, lameness, estrus detection, and electronic identification being the main research directions, and deep learning and machine learning represent the forefront of current research. Research hot topics mainly include social science in PLF, the environmental impact of PLF, information technology in PLF, and animal welfare in PLF. Future research in PLF should prioritize inter-institutional and inter-scholar communication and cooperation, integration of multidisciplinary and multimethod research approaches, and utilization of deep learning and machine learning. Furthermore, social science issues should be given due attention in PLF, and the integration of intelligent technologies in animal management should be strengthened, with a focus on animal welfare and the environmental impact of animal husbandry, to promote its sustainable development.

Research on Six-Axis Sensor-Based Step-Counting Algorithm for Grazing Sheep

Step counting is an effective method to assess the activity level of grazing sheep. However, existing step-counting algorithms have limited adaptability to sheep walking patterns and fail to eliminate false step counts caused by abnormal behaviors. Therefore, this study proposed a step-counting algorithm based on behavior classification designed explicitly for grazing sheep. The algorithm utilized regional peak detection and peak-to-valley difference detection to identify running and leg-shaking behaviors in sheep. It distinguished leg shaking from brisk walking behaviors through variance feature analysis. Based on the recognition results, different step-counting strategies were employed. When running behavior was detected, the algorithm divided the sampling window by the baseline step frequency and multiplied it by a scaling factor to accurately calculate the number of steps for running. No step counting was performed for leg-shaking behavior. For other behaviors, such as slow and brisk walking, a window peak detection algorithm was used for step counting. Experimental results demonstrate a significant improvement in the accuracy of the proposed algorithm compared to the peak detection-based method. In addition, the experimental results demonstrated that the average calculation error of the proposed algorithm in this study was 6.244%, while the average error of the peak detection-based step-counting algorithm was 17.556%. This indicates a significant improvement in the accuracy of the proposed algorithm compared to the peak detection method.

Uncovering Patterns in Dairy Cow Behaviour: A Deep Learning Approach with Tri-Axial Accelerometer Data

The accurate detection of behavioural changes represents a promising method of detecting the early onset of disease in dairy cows. This study assessed the performance of deep learning (DL) in classifying dairy cows' behaviour from accelerometry data acquired by single sensors on the cows' left flanks and compared the results with those obtained through classical machine learning (ML) from the same raw data. Twelve cows with a tri-axial accelerometer were observed for 136 ± 29 min each to detect five main behaviours: standing still, moving, feeding, ruminating and resting. For each 8 s time interval, 15 metrics were calculated, obtaining a dataset of 211,720 observation units and 15 columns. The entire dataset was randomly split into training (80%) and testing (20%) datasets. The DL accuracy, precision and sensitivity/recall were calculated and compared with the performance of classical ML models. The best predictive model was an 8-layer convolutional neural network (CNN) with an overall accuracy and F1 score equal to 0.96. The precision, sensitivity/recall and F1 score of single behaviours had the following ranges: 0.93-0.99. The CNN outperformed all the classical ML algorithms. The CNN used to monitor the cows' conditions showed an overall high performance in successfully predicting multiple behaviours using a single accelerometer.

Impact of sensor data pre-processing strategies and selection of machine learning algorithm on the prediction of metritis events in dairy cattle

With all the sensor data currently generated at high frequency in dairy farms, there is potential for earlier diagnosis of postpartum diseases compared with traditional monitoring methodologies. Our objectives were 1) to compare the impact of sensor data pre-processing on classifier performance by using multiple time windows before a given metritis event, while considering other cow-level factors and farm-scheduled activities; 2) to compare the performance of random forest (RF), k-nearest neighbors (k-NN), and support vector machine (SVM) classifiers at different decision thresholds using different number of past observations (time-lags) for the detection of behavioral patterns associated with changes in metritis scores; and 3) to compare classifier performance between each one of the five behaviors registered every hour by an ear-tag 3-axis accelerometer (CowManager, Agis Autimatisering, Harmelen, Netherlands). A total of 239 metritis events were created by comparing metritis scores between two consecutive clinical evaluations from cows that were retrospectively selected from a dataset containing sensor data and health information during the first 21 days postpartum from June 2014 to May 2017. Hourly sensor data classified by the accelerometer as either ruminating, eating, not active (including both standing or lying), and two different levels of activity (active and high activity) behaviors corresponding to the 3 days before each metritis event were aggregated every 24-, 12-, 6-, and 3-hour time windows. Multiple time-lags were also used to determine the optimal number of past observations needed for optimal classification. Similarly, different decision thresholds were compared in terms of model performance. Depending on the classifier, algorithm hyperparameters were optimized using grid search (RF, k-NN, SVM) and random search (RF). All behaviors changed throughout the study period and showed distinct daily patterns. From the three algorithms, RF had the highest F₁ score followed by k-NN and SVM. Furthermore, sensor data aggregated every 6- or 12-h time windows had the best model performance at multiple time-lags. We concluded that the data from the first 3 days post-partum should be discarded when studying metritis, and either one of the five behaviors measured with CowManager could be used when predicting metritis when sensor data were aggregated every 6- or 12-hour time windows, and using time-lags corresponding to 2-3 days before a given event, depending on the time window used. This study shows how to maximize sensor data in their potential for disease prediction, enhancing the performance of algorithms used in machine learning.

Classification of cow behavior patterns using inertial measurement units and a fully convolutional network model

In this study, we aimed to classify 7 cow behavior patterns automatically with an inertial measurement unit (IMU) using a fully convolutional network (FCN) algorithm. Behavioral data of 12 cows were collected by attaching an IMU in a waterproof box on the neck behind the head of each cow. Seven behavior patterns were considered: rub scratching (leg), ruminating-lying, lying, feeding, self-licking, rub scratching (neck), and social licking. To simplify the data and compare classification performance with or without magnetometer data, the 9-axis IMU data were reduced using the square root of the sum of squares to develop 2 datasets. Comparing the classification accuracy of the 3 models using a window size of 64 with 6-axis data and a window size of 128 with both 6-axis and 9-axis data, the best overall accuracy (83.75%) was achieved using the FCN model with a window size of 128 (12.8 s) using all IMU data. This model achieved classification accuracies of 83.2, 96.5, 92.8, 98.1, 82.9, 87.2, and 45.2% for ruminating-lying, lying, feeding, rub scratching (leg), self-licking, rub scratching (neck), and social licking, respectively. As a sequence of varied and intensive movement, the classification accuracy of behavior patterns related to skin disease was lower; better classification of these behavior patterns could be achieved with full IMU data and a larger window size. In the future, additional data will take into account different data types, such as audio and video data, to further enhance performance. In addition, an adaptive sliding window size will be used to improve model performance.

Behavioral Fingerprinting: Acceleration Sensors for Identifying Changes in Livestock Health

During disease or toxin challenges, the behavioral activities of grazing animals alter in response to adverse situations, potentially providing an indicator of their welfare status. Behavioral changes such as feeding behavior, rumination and physical behavior as well as expressive behavior, can serve as indicators of animal health and welfare. Sometimes behavioral changes are subtle and occur gradually, often missed by infrequent visual monitoring until the condition becomes acute. There is growing popularity in the use of sensors for monitoring animal health. Acceleration sensors have been designed to attach to ears, jaws, noses, collars and legs to detect the behavioral changes of cattle and sheep. So far, some automated acceleration sensors with high accuracies have been found to have the capacity to remotely monitor the behavioral patterns of cattle and sheep. These acceleration sensors have the potential to identify behavioral patterns of farm animals for monitoring changes in behavior which can indicate a deterioration in health. Here, we review the current automated accelerometer systems and the evidence they can detect behavioral patterns of animals for the application of potential directions and future solutions for automatically monitoring and the early detection of health concerns in grazing animals.

Identification of reindeer fine-scale foraging behaviour using tri-axial accelerometer data

Animal behavioural responses to the environment ultimately affect their survival. Monitoring animal fine-scale behaviour may improve understanding of animal functional response to the environment and provide an important indicator of the welfare of both wild and domesticated species. In this study, we illustrate the application of collar-attached acceleration sensors for investigating reindeer fine-scale behaviour. Using data from 19 reindeer, we tested the supervised machine learning algorithms Random forests, Support vector machines, and hidden Markov models to classify reindeer behaviour into seven classes: grazing, browsing low from shrubs or browsing high from trees, inactivity, walking, trotting, and other behaviours. We implemented leave-one-subject-out cross-validation to assess generalizable results on new individuals. Our main results illustrated that hidden Markov models were able to classify collar-attached accelerometer data into all our pre-defined behaviours of reindeer with reasonable accuracy while Random forests and Support vector machines were biased towards dominant classes. Random forests using 5-s windows had the highest overall accuracy (85%), while hidden Markov models were able to best predict individual behaviours and handle rare behaviours such as trotting and browsing high. We conclude that hidden Markov models provide a useful tool to remotely monitor reindeer and potentially other large herbivore species behaviour. These methods will allow us to quantify fine-scale behavioural processes in relation to environmental events.

Feature Selection Model Based on IWOA for Behavior Identification of Chicken

In order to reduce the influence of redundant features on the performance of the model in the process of accelerometer behavior recognition, and to improve the recognition accuracy of the model, this paper proposes an improved Whale Optimization algorithm with mixed strategy (IWOA) combined with the extreme gradient boosting algorithm (XGBoost) as a preferred method for chicken behavior identification features. A nine-axis inertial sensor was used to obtain the chicken behavior data. After noise reduction, the sliding window was used to extract 44 dimensional features in the time domain and frequency domain. To improve the search ability of the Whale Optimization algorithm for optimal solutions, the introduction of the good point set improves population diversity and expands the search range; the introduction of adaptive weight balances the search ability of the optimal solution in the early and late stages; the introduction of dimension-by-dimension lens imaging learning based on the adaptive weight factor perturbs the optimal solution and enhances the ability to jump out of the local optimal solution. This method's effectiveness was verified by recognizing cage breeders' feeding and drinking behaviors. The results show that the number of feature dimensions is reduced by 72.73%. At the same time, the behavior recognition accuracy is increased by 2.41% compared with the original behavior feature dataset, which is 95.58%. Compared with other dimensionality reduction methods, the IWOA-XGBoost model proposed in this paper has the highest recognition accuracy. The dimension reduction results have a certain degree of universality for different classification algorithms. This provides a method for behavior recognition based on acceleration sensor data.

High Precision Classification of Resting and Eating Behaviors of Cattle by Using a Collar-Fitted Triaxial Accelerometer Sensor

Cattle are less active than humans. Hence, it was hypothesized in this study that transmitting acceleration signals at a 1 min sampling interval to reduce storage load has the potential to improve the performance of motion sensors without affecting the precision of behavior classification. The behavior classification performance in terms of precision, sensitivity, and the F1-score of the 1 min serial datasets segmented in 3, 4, and 5 min window sizes based on nine algorithms were determined. The collar-fitted triaxial accelerometer sensor was attached on the right side of the neck of the two fattening Korean steers (age: 20 months) and the steers were observed for 6 h on day one, 10 h on day two, and 7 h on day three. The acceleration signals and visual observations were time synchronized and analyzed based on the objectives. The resting behavior was most correctly classified using the combination of a 4 min window size and the long short-term memory (LSTM) algorithm which resulted in 89% high precision, 81% high sensitivity, and 85% high F1-score. High classification performance (79% precision, 88% sensitivity, and 83% F1-score) was also obtained in classifying the eating behavior using the same classification method (4 min window size and an LSTM algorithm). The most poorly classified behavior was the active behavior. This study showed that the collar-fitted triaxial sensor measuring 1 min serial signals could be used as a tool for detecting the resting and eating behaviors of cattle in high precision by segmenting the acceleration signals in a 4 min window size and by using the LSTM classification algorithm.

A robust classification system for Southern Yellow cow behavior using 3-DoF accelerometers

Modern methods of monitoring help cow farmers save significantly monitoring time and improve cow health care efficiency. Behavioral changes when cows are sick may include increased or decreased daily activities such as increased lying or decreased walking time. Accelerometer advantages are low power consumption, small size, and lightweight. Thus, accelerometers have been widely used to monitor cow behavior. A cow monitoring system usually includes a central processor for receiving and processing information according to a behavioral classification algorithm through the cows’ movements. This paper introduces an effective classification system for Southern Yellow cow behavior using three degrees of freedom (3-DoF) accelerometers. The proposed classifier applied GBDT algorithm (16 seconds window) with five features, offers the good performance while investigating with four Southern Yellow cattle. The classification achievement was assessed and compared to existing ones regarding sensitivity, accuracy, and positive predictive value.

Development of a New Wearable 3D Sensor Node and Innovative Open Classification System for Dairy Cows’ Behavior

Simple Summary

In order to keep dairy cows under satisfactory health and welfare conditions, it is very important to monitor the animals in their living environment. With the support of technology, and, in particular, with the installation of sensors on neck-collars, cow behavior can be adequately monitored, and different behavioral patterns can be classified. In this study, an open and customizable device has been developed to classify the behaviors of dairy cows. The device communicates with a mobile application via Bluetooth to acquire raw data from behavioral observations and via an ad hoc radio channel to send the data from the device to the gateway. After observing 32 cows on 3 farms for a total of 108 h, several machine learning algorithms were trained to classify their behaviors. The decision tree algorithm was found to be the best compromise between complexity and accuracy to classify standing, lying, eating, and ruminating. The open nature of the system enables the addition of other functions (e.g., localization) and the integration with other information sources, e.g., climatic sensors, to provide a more complete picture of cow health and welfare in the barn.

Abstract

Monitoring dairy cattle behavior can improve the detection of health and welfare issues for early interventions. Often commercial sensors do not provide researchers with sufficient raw and open data; therefore, the aim of this study was to develop an open and customizable system to classify cattle behaviors. A 3D accelerometer device and host-board (i.e., sensor node) were embedded in a case and fixed on a dairy cow collar. It was developed to work in two modes: (1) acquisition mode, where a mobile application supported the raw data collection during observations; and (2) operating mode, where data was processed and sent to a gateway and on the cloud. Accelerations were sampled at 25 Hz and behaviors were classified in 10-min windows. Several algorithms were trained with the 108 h of behavioral data acquired from 32 cows on 3 farms, and after evaluating their computational/memory complexity and accuracy, the Decision Tree algorithm was selected. This model detected standing, lying, eating, and ruminating with an average accuracy of 85.12%. The open nature of this system enables for the addition of other functions (e.g., real-time localization of cows) and the integration with other information sources, e.g., microenvironment and air quality sensors, thereby enhancing data processing potential.

Behavioural Classification of Cattle Using Neck-Mounted Accelerometer-Equipped Collars

Monitoring and classification of dairy cattle behaviours is essential for optimising milk yields. Early detection of illness, days before the critical conditions occur, together with automatic detection of the onset of oestrus cycles is crucial for obviating prolonged cattle treatments and improving the pregnancy rates. Accelerometer-based sensor systems are becoming increasingly popular, as they are automatically providing information about key cattle behaviours such as the level of restlessness and the time spent ruminating and eating, proxy measurements that indicate the onset of heat events and overall welfare, at an individual animal level. This paper reports on an approach to the development of algorithms that classify key cattle states based on a systematic dimensionality reduction process through two feature selection techniques. These are based on Mutual Information and Backward Feature Elimination and applied on knowledge-specific and generic time-series extracted from raw accelerometer data. The extracted features are then used to train classification models based on a Hidden Markov Model, Linear Discriminant Analysis and Partial Least Squares Discriminant Analysis. The proposed feature engineering methodology permits model deployment within the computing and memory restrictions imposed by operational settings. The models were based on measurement data from 18 steers, each animal equipped with an accelerometer-based neck-mounted collar and muzzle-mounted halter, the latter providing the truthing data. A total of 42 time-series features were initially extracted and the trade-off between model performance, computational complexity and memory footprint was explored. Results show that the classification model that best balances performance and computation complexity is based on Linear Discriminant Analysis using features selected through Backward Feature Elimination. The final model requires 1.83 ± 1.00 ms to perform feature extraction with 0.05 ± 0.01 ms for inference with an overall balanced accuracy of 0.83.

Prediction of Cow Calving in Extensive Livestock Using a New Neck-Mounted Sensorized Wearable Device: A Pilot Study

In this study, new low-cost neck-mounted sensorized wearable device is presented to help farmers detect the onset of calving in extensive livestock farming by continuously monitoring cow data. The device incorporates three sensors: an inertial measurement unit (IMU), a global navigation satellite system (GNSS) receiver, and a thermometer. The hypothesis of this study was that onset calving is detectable through the analyses of the number of transitions between lying and standing of the animal (lying bouts). A new algorithm was developed to detect calving, analysing the frequency and duration of lying and standing postures. An important novelty is that the proposed algorithm has been designed with the aim of being executed in the embedded microcontroller housed in the cow’s collar and, therefore, it requires minimal computational resources while allowing for real time data processing. In this preliminary study, six cows were monitored during different stages of gestation (before, during, and after calving), both with the sensorized wearable device and by human observers. It was carried out on an extensive livestock farm in Salamanca (Spain), during the period from August 2020 to July 2021. The preliminary results obtained indicate that lying-standing animal states and transitions may be useful to predict calving. Further research, with data obtained in future calving of cows, is required to refine the algorithm.

Strategy to Predict High and Low Frequency Behaviors Using Triaxial Accelerometers in Grazing of Beef Cattle

Simple Summary

Monitoring animal activity in production systems is an important tool for obtaining information on health, production, and reproduction. In this study, we evaluated the use of accelerometers with different strategies to predict the grazing behavior of Nelore cattle. This research was conducted in an environment both more challenging and representative of the practices adopted in livestock production systems in Brazil. The results of this study showed that the use of the Random Forest algorithm, together with techniques for resampling the training data of the models, classified the studied behaviors with high accuracy, especially for important, and less frequent activities such as water consumption frequency.

Abstract

Knowledge of animal behavior can be indicative of the well-being, health, productivity, and reproduction of animals. The use of accelerometers to classify and predict animal behavior can be a tool for continuous animal monitoring. Therefore, the aim of this study was to provide strategies for predicting more and less frequent beef cattle grazing behaviors. The behavior activities observed were grazing, ruminating, idle, water consumption frequency (WCF), feeding (supplementation) and walking. Three Machine Learning algorithms: Random Forest (RF), Support Vector Machine (SVM) and Naïve Bayes Classifier (NBC) and two resample methods: under and over-sampling, were tested. Overall accuracy was higher for RF models trained with the over-sampled dataset. The greatest sensitivity (0.808) for the less frequent behavior (WCF) was observed in the RF algorithm trained with the under-sampled data. The SVM models only performed efficiently when classifying the most frequent behavior (idle). The greatest predictor in the NBC algorithm was for ruminating behavior, with the over-sampled training dataset. The results showed that the behaviors of the studied animals were classified with high accuracy and specificity when the RF algorithm trained with the resampling methods was used. Resampling training datasets is a strategy to be considered, especially when less frequent behaviors are of interest.

Intelligent Perception-Based Cattle Lameness Detection and Behaviour Recognition: A Review

Simple Summary

Cattle lameness detection as well as behaviour recognition are the two main objectives in the applications of precision livestock farming (PLF). Over the last five years, the development of smart sensors, big data, and artificial intelligence has offered more automatic tools. In this review, we discuss over 100 papers that used automated techniques to detect cattle lameness and to recognise animal behaviours. To assist researchers and policy-makers in promoting various livestock technologies for monitoring cattle welfare and productivity, we conducted a comprehensive investigation of intelligent perception for cattle lameness detection and behaviour analysis in the PLF domain. Based on the literature review, we anticipate that PLF will develop in an objective, autonomous, and real-time direction. Additionally, we suggest that further research should be dedicated to improving the data quality, modeling accuracy, and commercial availability.

Abstract

The growing world population has increased the demand for animal-sourced protein. However, animal farming productivity is faced with challenges from traditional farming practices, socioeconomic status, and climate change. In recent years, smart sensors, big data, and deep learning have been applied to animal welfare measurement and livestock farming applications, including behaviour recognition and health monitoring. In order to facilitate research in this area, this review summarises and analyses some main techniques used in smart livestock farming, focusing on those related to cattle lameness detection and behaviour recognition. In this study, more than 100 relevant papers on cattle lameness detection and behaviour recognition have been evaluated and discussed. Based on a review and a comparison of recent technologies and methods, we anticipate that intelligent perception for cattle behaviour and welfare monitoring will develop towards standardisation, a larger scale, and intelligence, combined with Internet of things (IoT) and deep learning technologies. In addition, the key challenges and opportunities of future research are also highlighted and discussed.

Machine Learning to Detect Posture and Behavior in Dairy Cows: Information from an Accelerometer on the Animal's Left Flank

Simple Summary

This study analyzed the possibility of automatically detecting dairy cow behavior by combining the use of a single triaxial accelerometer applied to the animal’s left flank with a machine learning technique. This combination enabled the detection of posture and the main types of behavior that are extremely useful in evaluating the animal’s welfare and health such as resting, feeding, and rumination with a high degree of accuracy. The novelty of the study was the success in reaching a high accuracy in detecting five different behaviors and the animal posture by using a single sensor and allowing farmers to save money. To the best of our knowledge, this is the first study that has successfully explored the feasibility of locating a sensor on the animal’s left flank, showing the opportunity of automatically measuring some physiological parameters, such as those ones related to respiration and rumen health, in a non-invasive way.

Abstract

The aim of the present study was to develop a model to identify posture and behavior from data collected by a triaxial accelerometer located on the left flank of dairy cows and evaluate its accuracy and precision. Twelve Italian Red-and-White lactating cows were equipped with an accelerometer and observed on average for 136 ± 29 min per cow by two trained operators as a reference. The acceleration data were grouped in time windows of 8 s overlapping by 33.0%, for a total of 35,133 rows. For each row, 32 different features were extracted and used by machine learning algorithms for the classification of posture and behavior. To build up a predictive model, the dataset was split in training and testing datasets, characterized by 75.0 and 25.0% of the observations, respectively. Four algorithms were tested: Random Forest, K Nearest Neighbors, Extreme Boosting Algorithm (XGB), and Support Vector Machine. The XGB model showed the best accuracy (0.99) and Cohen’s kappa (0.99) in predicting posture, whereas the Random Forest model had the highest overall accuracy in predicting behaviors (0.76), showing a balanced accuracy from 0.96 for resting to 0.77 for moving. Overall, very accurate detection of the posture and resting behavior were achieved.

A Comprehensive Overview of Technologies for Species and Habitat Monitoring and Conservation

The range of technologies currently used in biodiversity conservation is staggering, with innovative uses often adopted from other disciplines and being trialed in the field. We provide the first comprehensive overview of the current (2020) landscape of conservation technology, encompassing technologies for monitoring wildlife and habitats, as well as for on-the-ground conservation management (e.g., fighting illegal activities). We cover both established technologies (routinely deployed in conservation, backed by substantial field experience and scientific literature) and novel technologies or technology applications (typically at trial stage, only recently used in conservation), providing examples of conservation applications for both types. We describe technologies that deploy sensors that are fixed or portable, attached to vehicles (terrestrial, aquatic, or airborne) or to animals (biologging), complemented with a section on wildlife tracking. The last two sections cover actuators and computing (including web platforms, algorithms, and artificial intelligence).

Classifying Goliath Grouper (Epinephelus itajara) Behaviors from a Novel, Multi-Sensor Tag

Inertial measurement unit sensors (IMU; i.e., accelerometer, gyroscope and magnetometer combinations) are frequently fitted to animals to better understand their activity patterns and energy expenditure. Capable of recording hundreds of data points a second, these sensors can quickly produce large datasets that require methods to automate behavioral classification. Here, we describe behaviors derived from a custom-built multi-sensor bio-logging tag attached to Atlantic Goliath grouper (Epinephelus itajara) within a simulated ecosystem. We then compared the performance of two commonly applied machine learning approaches (random forest and support vector machine) to a deep learning approach (convolutional neural network, or CNN) for classifying IMU data from this tag. CNNs are frequently used to recognize activities from IMU data obtained from humans but are less commonly considered for other animals. Thirteen behavioral classes were identified during ethogram development, nine of which were classified. For the conventional machine learning approaches, 187 summary statistics were extracted from the data, including time and frequency domain features. The CNN was fed absolute values obtained from fast Fourier transformations of the raw tri-axial accelerometer, gyroscope and magnetometer channels, with a frequency resolution of 512 data points. Five metrics were used to assess classifier performance; the deep learning approach performed better across all metrics (Sensitivity = 0.962; Specificity = 0.996; F₁-score = 0.962; Matthew’s Correlation Coefficient = 0.959; Cohen’s Kappa = 0.833) than both conventional machine learning approaches. Generally, the random forest performed better than the support vector machine. In some instances, a conventional learning approach yielded a higher performance metric for particular classes (e.g., the random forest had a F₁-score of 0.971 for backward swimming compared to 0.955 for the CNN). Deep learning approaches could potentially improve behavioral classification from IMU data, beyond that obtained from conventional machine learning methods.

Training and Validating a Machine Learning Model for the Sensor-Based Monitoring of Lying Behavior in Dairy Cows on Pasture and in the Barn

Simple Summary

There are various systems available for health monitoring and heat detection in dairy cows. By continuously monitoring different behavioral patterns (e.g., lying, ruminating, and feeding), these systems detect behavioral changes linked to health disorders and estrous. Most of the systems were developed for cows kept indoors, and only a few systems are available for pasture-based farms. The systems developed for the barn failed to detect the targeted behavior and thereby its changes on the pasture and vice versa. Therefore, our goal was to train and validate a machine learning model for the automated prediction of lying behavior in dairy cows kept on pastures, as well as indoors. Data collection was conducted on three dairy farms where cows were equipped with the collar-based prototype of the monitoring system and recorded with cameras in parallel. The derived dataset was used to develop the machine learning model. The model performed well in predicting lying behavior in dairy cows both on the pasture and in the barn. Therefore, the building of the model presents a successful first step towards the development of a monitoring system for dairy cows kept on pasture and in the barn.

Abstract

Monitoring systems assist farmers in monitoring the health of dairy cows by predicting behavioral patterns (e.g., lying) and their changes with machine learning models. However, the available systems were developed either for indoors or for pasture and fail to predict the behavior in other locations. Therefore, the goal of our study was to train and evaluate a model for the prediction of lying on a pasture and in the barn. On three farms, 7–11 dairy cows each were equipped with the prototype of the monitoring system containing an accelerometer, a magnetometer and a gyroscope. Video observations on the pasture and in the barn provided ground truth data. We used 34.5 h of datasets from pasture for training and 480.5 h from both locations for evaluating. In comparison, random forest, an orientation-independent feature set with 5 s windows without overlap, achieved the highest accuracy. Sensitivity, specificity and accuracy were 95.6%, 80.5% and 87.4%, respectively. Accuracy on the pasture (93.2%) exceeded accuracy in the barn (81.4%). Ruminating while standing was the most confused with lying. Out of individual lying bouts, 95.6 and 93.4% were identified on the pasture and in the barn, respectively. Adding a model for standing up events and lying down events could improve the prediction of lying in the barn.

Using Machine Learning for Remote Behaviour Classification-Verifying Acceleration Data to Infer Feeding Events in Free-Ranging Cheetahs

Behavioural studies of elusive wildlife species are challenging but important when they are threatened and involved in human-wildlife conflicts. Accelerometers (ACCs) and supervised machine learning algorithms (MLAs) are valuable tools to remotely determine behaviours. Here we used five captive cheetahs in Namibia to test the applicability of ACC data in identifying six behaviours by using six MLAs on data we ground-truthed by direct observations. We included two ensemble learning approaches and a probability threshold to improve prediction accuracy. We used the model to then identify the behaviours in four free-ranging cheetah males. Feeding behaviours identified by the model and matched with corresponding GPS clusters were verified with previously identified kill sites in the field. The MLAs and the two ensemble learning approaches in the captive cheetahs achieved precision (recall) ranging from 80.1% to 100.0% (87.3% to 99.2%) for resting, walking and trotting/running behaviour, from 74.4% to 81.6% (54.8% and 82.4%) for feeding behaviour and from 0.0% to 97.1% (0.0% and 56.2%) for drinking and grooming behaviour. The model application to the ACC data of the free-ranging cheetahs successfully identified all nine kill sites and 17 of the 18 feeding events of the two brother groups. We demonstrated that our behavioural model reliably detects feeding events of free-ranging cheetahs. This has useful applications for the determination of cheetah kill sites and helping to mitigate human-cheetah conflicts.

Precision Agriculture for Crop and Livestock Farming-Brief Review

In the last few decades, agriculture has played an important role in the worldwide economy. The need to produce more food for a rapidly growing population is creating pressure on crop and animal production and a negative impact to the environment. On the other hand, smart farming technologies are becoming increasingly common in modern agriculture to assist in optimizing agricultural and livestock production and minimizing the wastes and costs. Precision agriculture (PA) is a technology-enabled, data-driven approach to farming management that observes, measures, and analyzes the needs of individual fields and crops. Precision livestock farming (PLF), relying on the automatic monitoring of individual animals, is used for animal growth, milk production, and the detection of diseases as well as to monitor animal behavior and their physical environment, among others. This study aims to briefly review recent scientific and technological trends in PA and their application in crop and livestock farming, serving as a simple research guide for the researcher and farmer in the application of technology to agriculture. The development and operation of PA applications involve several steps and techniques that need to be investigated further to make the developed systems accurate and implementable in commercial environments.

A prediction and imputation method for marine animal movement data

Data prediction and imputation are important parts of marine animal movement trajectory analysis as they can help researchers understand animal movement patterns and address missing data issues. Compared with traditional methods, deep learning methods can usually provide enhanced pattern extraction capabilities, but their applications in marine data analysis are still limited. In this research, we propose a composite deep learning model to improve the accuracy of marine animal trajectory prediction and imputation. The model extracts patterns from the trajectories with an encoder network and reconstructs the trajectories using these patterns with a decoder network. We use attention mechanisms to highlight certain extracted patterns as well for the decoder. We also feed these patterns into a second decoder for prediction and imputation. Therefore, our approach is a coupling of unsupervised learning with the encoder and the first decoder and supervised learning with the encoder and the second decoder. Experimental results demonstrate that our approach can reduce errors by at least 10% on average comparing with other methods.

Quantifying finer-scale behaviours using self-organising maps (SOMs) to link accelerometery signatures with behavioural patterns in free-roaming terrestrial animals

Collecting quantitative information on animal behaviours is difficult, especially from cryptic species or species that alter natural behaviours under observation. Using harness-mounted tri-axial accelerometers free-roaming domestic cats (Felis Catus) we developed a methodology that can precisely classify finer-scale behaviours. We further tested the effect of a prey-protector device designed to reduce prey capture. We aligned accelerometer traces collected at 50 Hz with video files (60 fps) and labelled 12 individual behaviours, then trained a supervised machine-learning algorithm using Kohonen super self-organising maps (SOM). The SOM was able to predict individual behaviours with a ~ 99.6% overall accuracy, which was slightly better than for random forest estimates using the same dataset (98.9%). There was a significant effect of sample size, with precision and sensitivity decreasing rapidly below 2000 1-s observations. We were also able to detect a behaviour specific reduction in the predictability when cats were fitted with the prey-protector device indicating it altered biomechanical gait. Our results can be applied in movement ecology, zoology and conservation, where habitat specific movement performance between predators or prey may be critical to managing species of conservation significance, or in veterinary and agricultural fields, where early detection of movement pathologies can improve animal welfare.

Evaluation of an electronic system for monitoring dairy cow rumination in a grazing-based system

This research aimed to compare the rumination time estimated by an electronic monitoring system with direct visual observations of the rumination of dairy cows in a grazing-based system. Also, it aimed to estimate the agreement between different time intervals of the direct visual observations. To that end, we have observed twelve lactating cows for 16 h a day for 11 days. Then, we compared the data provided by the HealthyCow24® software with the visual observations to use as reference. After that, we estimated the Pearson's correlation coefficients, the linear regression, and the Bland-Altman plot by using the SAS software. Results showed that the rumination data estimated by the HealthyCow24® software and the visual observations were highly correlated (0.81; P < 0.0001). In the Bland-Altman analysis, we observed that the average of the standard deviations between the visual observation and the electronic monitoring system was - 2.14 min during a 2-h period. The upper limit (95%) was 30.61 min/2 h and the lower limit (95%) was - 34.88 min/2 h. Moreover, visual observations at intervals of up to 15 min were correlated with the data estimated by the electronic monitoring system and observed at 5-min intervals. In conclusion, the Allflex SCR electronic monitoring system is efficient in measuring the rumination time of grazing cows. Also, direct visual observations with intervals no longer than 15 min are reliable when used for evaluating the behavior of cows without losing data accuracy.

Sensor and Video: Two Complementary Approaches for Evaluation of Dairy Cow Behavior after Calving Sensor Attachment

Studies evaluating calving sensors provided evidence that attaching the sensor to the tail may lead to changes in the cows' behavior. Two different calving sensors were attached to 18 cows, all of which were equipped with a rumen bolus to record their activity. Two methodological approaches were applied to detect potential behavioral changes: analysis of homogeneity of variance in cow activity (5 days pre-sensor and 24 h post-sensor) and analysis of video-recorded behavior (12 h pre- and post-sensor, respectively) in a subgroup. The average results across the sample showed no significant changes in the variability of activity and no statistically significant mean differences in most visually analyzed behaviors, namely walking, eating, drinking, social interaction, tail raising, rubbing the tail, and the number of standing and lying bouts after calving sensor attachment. In addition to considering mean values across all cows, individual cow investigations revealed an increased number of time slots showing a significant increase in the variability of activity and an increased frequency of tail raising and rubbing the tail on objects after calving sensor attachment in some cows, which should be investigated in more detail on a larger scale.

Classification of Cattle Behaviours Using Neck-Mounted Accelerometer-Equipped Collars and Convolutional Neural Networks

Monitoring cattle behaviour is core to the early detection of health and welfare issues and to optimise the fertility of large herds. Accelerometer-based sensor systems that provide activity profiles are now used extensively on commercial farms and have evolved to identify behaviours such as the time spent ruminating and eating at an individual animal level. Acquiring this information at scale is central to informing on-farm management decisions. The paper presents the development of a Convolutional Neural Network (CNN) that classifies cattle behavioural states ('rumination', 'eating' and 'other') using data generated from neck-mounted accelerometer collars. During three farm trials in the United Kingdom (Easter Howgate Farm, Edinburgh, UK), 18 steers were monitored to provide raw acceleration measurements, with ground truth data provided by muzzle-mounted pressure sensor halters. A range of neural network architectures are explored and rigorous hyper-parameter searches are performed to optimise the network. The computational complexity and memory footprint of CNN models are not readily compatible with deployment on low-power processors which are both memory and energy constrained. Thus, progressive reductions of the CNN were executed with minimal loss of performance in order to address the practical implementation challenges, defining the trade-off between model performance versus computation complexity and memory footprint to permit deployment on micro-controller architectures. The proposed methodology achieves a compression of 14.30 compared to the unpruned architecture but is nevertheless able to accurately classify cattle behaviours with an overall F1 score of 0.82 for both FP32 and FP16 precision while achieving a reasonable battery lifetime in excess of 5.7 years.

Caution is warranted when using animal space-use and movement to infer behavioral states

BACKGROUND

Identifying the behavioral state for wild animals that can't be directly observed is of growing interest to the ecological community. Advances in telemetry technology and statistical methodologies allow researchers to use space-use and movement metrics to infer the underlying, latent, behavioral state of an animal without direct observations. For example, researchers studying ungulate ecology have started using these methods to quantify behaviors related to mating strategies. However, little work has been done to determine if assumed behaviors inferred from movement and space-use patterns correspond to actual behaviors of individuals.

METHODS

Using a dataset with male and female white-tailed deer location data, we evaluated the ability of these two methods to correctly identify male-female interaction events (MFIEs). We identified MFIEs using the proximity of their locations in space as indicators of when mating could have occurred. We then tested the ability of utilization distributions (UDs) and hidden Markov models (HMMs) rendered with single sex location data to identify these events.

RESULTS

For white-tailed deer, male and female space-use and movement behavior did not vary consistently when with a potential mate. There was no evidence that a probability contour threshold based on UD volume applied to an individual's UD could be used to identify MFIEs. Additionally, HMMs were unable to identify MFIEs, as single MFIEs were often split across multiple states and the primary state of each MFIE was not consistent across events.

CONCLUSIONS

Caution is warranted when interpreting behavioral insights rendered from statistical models applied to location data, particularly when there is no form of validation data. For these models to detect latent behaviors, the individual needs to exhibit a consistently different type of space-use and movement when engaged in the behavior. Unvalidated assumptions about that relationship may lead to incorrect inference about mating strategies or other behaviors.

Predicting feed intake using modelling based on feeding behaviour in finishing beef steers

Current techniques for measuring feed intake in housed cattle are both expensive and time-consuming making them unsuitable for use on commercial farms. Estimates of individual animal intake are required for assessing production efficiency. The aim of this study was to predict individual animal intake using parameters that can be easily obtained on commercial farms including feeding behaviour, liveweight and age. In total, 80 steers were used, and each steer was allocated to one of two diets (40 per diet) which consisted of (g/kg; DM) forage to concentrate ratios of either 494:506 (MIXED) or 80:920 (CONC). Individual daily fresh weight intakes (FWI; kg/day) were recorded for each animal using 32 electronic feeders over a 56-day period, and individual DM intakes (DMI; kg/day) subsequently calculated. Individual feeding behaviour variables were calculated for each day of the measurement period from the electronic feeders and included: total number of visits to the feeder, total time spent at the feeder (TOTFEEDTIME), total time where feed was consumed (TIMEWITHFEED) and average length of time during each visit to the feeder. These feeding behaviour variables were chosen due to ease of obtaining from accelerometers. Four modelling techniques to predict individual animal intake were examined, based on (i) individual animal TOTFEEDTIME relative expressed as a proportion of the dietary group (GRP) and total GRP intake, (ii) multiple linear regression (REG) (iii) random forests (RF) and (iv) support vector regressor (SVR). Each model was used to predict CONC and MIXED diets separately, giving eight prediction models, (i) GRP_CONC, (ii) GRP_MIXED, (iii) REG_CONC, (iv) REG_MIXED, (v) RF_CONC, (vi) RF_MIXED, (vii) SVR_CONC and (viii) SVR_MIXED. Each model was tested on FWI and DMI. Model performance was assessed using repeated measures correlations (R²_RM) to capture the repeated nature of daily intakes compared with standard R², RMSE and mean absolute error (MAE). REG, RF and SVR models predicted FWI with R²_RM = 0.1–0.36, RMSE = 1.51–2.96 kg and MAE = 1.19–2.49 kg, and DMI with R²_RM = 0.13–0.19, RMSE = 1.15–1.61 kg and MAE = 0.9–1.28 kg. The GRP models predicted FWI with R²_RM = 0.42–0.49, RMSE = 2.76–3.88 kg and MAE = 2.46–3.47 kg, and DMI with R²_RM = 0.32–0.44, RMSE = 0.32–0.44 kg, MAE = 1.55–2.22 kg. Whilst more simplistic GRP models showed higher R²_RM than regression and machine learning techniques, these models had larger errors, likely due to individual feeding patterns not being captured. Although regression and machine learning techniques produced lower errors associated with individual intakes, overall precision of prediction was too low for practical use.

Classification of broiler behaviours using triaxial accelerometer and machine learning

Understanding broiler behaviours provides important implications for animal well-being and farm management. The objectives of this study were to classify specific broiler behaviours by analysing data from wearable accelerometers using two machine learning models, K-Nearest Neighbour (KNN) and Support Vector Machine (SVM). Lightweight triaxial accelerometers were used to record accelerations of nine 7-week-old broilers at a sampling frequency of 40 Hz. A total of 261.6-min data were labelled for four behaviours - walking, resting, feeding and drinking. Instantaneous motion features including magnitude area, vector magnitude, movement variation, energy, and entropy were extracted and stored in a dataset which was then segmented by one of the six window lengths (1, 3, 5, 7, 10 and 20 s) with 50% overlap between consecutive windows. The mean, variation, SD, minimum and maximum of each instantaneous motion feature and two-way correlations of acceleration data were calculated within each window, yielding a total of 43 statistic features for training and testing of machine learning models. Performance of the models was evaluated using pure behaviour datasets (single behaviour type per dataset) and continuous behaviour datasets (continuous recording that involved multiple behaviour types per dataset). For pure behaviour datasets, both KNN and SVM models showed high sensitivities in classifying broiler resting (87% and 85%, respectively) and walking (99% and 99%, respectively). The accuracies of SVM were higher than KNN in differentiating feeding (88% and 75%, respectively) and drinking (83% and 62%, respectively) behaviours. Sliding window with 1-s length yielded the best performance for classifying continuous behaviour datasets. The performance of classification model generally improved as more birds were included for training. In conclusion, classification of specific broiler behaviours can be achieved by recording bird triaxial accelerations and analysing acceleration data through machine learning. Performances of different machine learning models differ in classifying specific broiler behaviours.

A LoRa sensor network for monitoring pastured livestock location and activity

Precision technologies for confinement animal agricultural systems have increased rapidly over the past decade, though precision technology solutions for pastured livestock remain limited. There are a number of reasons for this limited expansion of technologies for pastured animals, including networking availability and reliability, power requirements, and expense, among others. The objective of this work was to demonstrate a rapidly deployable long-range radio (LoRa) based, low-cost sensor suite that can be used to track location and activity of pastured livestock. The sensor is comprised of an inexpensive Arduino-compatible microprocessor, a generic MPU-9250 motion sensor which contains a 3-axis accelerometer, 3-axis magnetometer, and a 3-axis gyroscope, a generic GPS receiver, and a RFM95W generic LoRa radio. The microprocessor can be programmed flexibly using the open source Arduino IDE software to adjust the frequency of sampling, the data packet to send, and what conditions are needed to operate. The LoRa radio transmits to a Dragino LoRa gateway which can also be flexibly programmed through the Arduino IDE software to send data to local storage or, in cases where a web or cellular connection is available, to cloud storage. The sensor was powered using a USB cord connected to a 3,350 mAh lithium-ion battery pack. The Dragino gateway was programmed to upload data to the ThingSpeak IoT application programming interface for data storage, handling, and visualization. Evaluations showed minimal benefit associated with reducing sampling frequency as a strategy to preserve battery life. Packet loss ranged from 40% to 60%. In a 3 d evaluation on pastured sheep, the sensor suite was able to report GPS locations, inertial sensor readings, and temperature. Preliminary demonstrations of our system are satisfactory to detect animal location based on GPS data in real-time. This system has clear utility as a lower-cost strategy to deploy flexible, useful precision technologies for pasture-based livestock species.

A Pilot Study Using Accelerometers to Characterise the Licking Behaviour of Penned Cattle at a Mineral Block Supplement

Simple Summary

Quantifying mineral block supplement intake by individual beef cattle is a challenging task but may enable improved efficiency of supplement use particularly in a grazed system. Estimating time spent licking when cattle access the mineral block supplement can be useful for predicting intake on an individual basis. The advancement of sensor technology has facilitated collection of individual data associated with ingestive behaviours such as feeding and licking duration. This experiment was intended to investigate the effectiveness of wearable tri-axial accelerometers fitted on both neck-collar and ear-tag to identify the licking behaviour of beef cattle by distinguishing it from eating, standing and lying behaviours. The capability of tri-axial accelerometers to classify licking behaviour in beef cattle revealed in this study would offer the possibility of measuring time spent licking and further developing a practical method of estimating mineral block supplement intake by individual grazing cattle.

Abstract

Identifying the licking behaviour in beef cattle may provide a means to measure time spent licking for estimating individual block supplement intake. This study aimed to determine the effectiveness of tri-axial accelerometers deployed in a neck-collar and an ear-tag, to characterise the licking behaviour of beef cattle in individual pens. Four, 2-year-old Angus steers weighing 368 ± 9.3 kg (mean ± SD) were used in a 14-day study. Four machine learning (ML) algorithms (decision trees [DT], random forest [RF], support vector machine [SVM] and k-nearest neighbour [kNN]) were employed to develop behaviour classification models using three different ethograms: (1) licking vs. eating vs. standing vs. lying; (2) licking vs. eating vs. inactive; and (3) licking vs. non-licking. Activities were video-recorded from 1000 to 1600 h daily when access to supplement was provided. The RF algorithm exhibited a superior performance in all ethograms across the two deployment modes with an overall accuracy ranging from 88% to 98%. The neck-collar accelerometers had a better performance than the ear-tag accelerometers across all ethograms with sensitivity and positive predictive value (PPV) ranging from 95% to 99% and 91% to 96%, respectively. Overall, the tri-axial accelerometer was capable of identifying licking behaviour of beef cattle in a controlled environment. Further research is required to test the model under actual grazing conditions.

Behaviour Classification on Giraffes (Giraffa camelopardalis) Using Machine Learning Algorithms on Triaxial Acceleration Data of Two Commonly Used GPS Devices and Its Possible Application for Their Management and Conservation

Averting today's loss of biodiversity and ecosystem services can be achieved through conservation efforts, especially of keystone species. Giraffes (Giraffa camelopardalis) play an important role in sustaining Africa's ecosystems, but are 'vulnerable' according to the IUCN Red List since 2016. Monitoring an animal's behavior in the wild helps to develop and assess their conservation management. One mechanism for remote tracking of wildlife behavior is to attach accelerometers to animals to record their body movement. We tested two different commercially available high-resolution accelerometers, e-obs and Africa Wildlife Tracking (AWT), attached to the top of the heads of three captive giraffes and analyzed the accuracy of automatic behavior classifications, focused on the Random Forests algorithm. For both accelerometers, behaviors of lower variety in head and neck movements could be better predicted (i.e., feeding above eye level, mean prediction accuracy e-obs/AWT: 97.6%/99.7%; drinking: 96.7%/97.0%) than those with a higher variety of body postures (such as standing: 90.7-91.0%/75.2-76.7%; rumination: 89.6-91.6%/53.5-86.5%). Nonetheless both devices come with limitations and especially the AWT needs technological adaptations before applying it on animals in the wild. Nevertheless, looking at the prediction results, both are promising accelerometers for behavioral classification of giraffes. Therefore, these devices when applied to free-ranging animals, in combination with GPS tracking, can contribute greatly to the conservation of giraffes.

Detection of cow hind-leg activity during milking by using a 3-dimensional accelerometer attached to the milking cluster

•

Hind-leg activity during milking in cows is a good welfare indicator

•

Cow hind-leg movements are transmitted to the milking claw

•

3D accelerometers attached to the milking claws accurately record cow hind-leg movements

Hind-leg activity in dairy cows during milking is considered an indicator of a stressful situation or discomfort. Automatic detection of cow hind-leg activity during milking could be used to monitor deficiencies at the cow, milking machine, or human level. We assumed that the milking cluster follows cow hind-leg movements during milking. Therefore, the aim of our study was to estimate the ability of a 3-dimensional (3D) accelerometer attached to the claw's hook of the milking cluster to indirectly measure cow hind-leg activity during milking. The behavior of 45 dairy cows during 1 morning and 1 evening milking in a 2 × 3 auto tandem milking parlor was monitored by using direct visual observations and a 3D accelerometer attached directly to the hook of each milking claw. We recorded when hind-leg foot-lifting or stepping (undifferentiated) and kicking behaviors occurred. Detected activities were obtained after applying a standard deviation filter (over a sliding window of 1 s) on the x (left/right lateral) acceleration axis only. We set the detection threshold at 0.13 g (i.e., 1.27 m·s⁻²), based on the visual evaluation of 10 “reference” graphs (from 5 randomly chosen cows). Periods of activity were created for each observed or detected activity (and for consecutive activities that were less than 3 s apart) to take into account the milking cluster's rocking motion resulting from a single movement. Observed and detected periods of activity were considered matching when they occurred simultaneously (with a 2-s delay acceptance in the visual observations). In total, 472 periods of hind-leg activity were observed. The detection of hind-leg activity by using a 3D accelerometer attached to the milking cluster was found to have 69% sensitivity, 99% specificity, 75% positive predictive value, 98% negative predictive value, and an overall accuracy of 97%. On average, cows were observed being active 0.94 times per minute and being detected active 0.86 times per minute. The observed and detected periods of activity were found to be correlated (r = 0.657). Therefore, the use of a 3D accelerometer attached to the milking cluster seems promising for reliably estimating cow hind-leg activity during milking. Nevertheless, further investigations are necessary to clarify whether other factors can cause variations in milking cluster movements, regardless of cow hind-leg activity, and to generalize our method to better identify deficiencies in cow–machine–human interactions.

Prediction of lameness using automatically recorded activity, behavior and production data in post-parturient Irish dairy cows

BACKGROUND

Although visual locomotion scoring is inexpensive and simplistic, it is also time consuming and subjective. Automated lameness detection methods have been developed to replace the visual locomotion scoring and aid in early and accurate detection. Several types of sensors are measuring traits such as activity, lying behavior or temperature. Previous studies on automatic lameness detection have been unable to achieve high accuracy in combination with practical implementation in a on farm commercial setting. The objective of our research was to develop a prediction model for lameness in dairy cattle using a combination of remote sensor technology and other animal records that will translate sensor data into easy to interpret classified locomotion information for the farmer. During an 11-month period, data from 164 Holstein-Friesian dairy cows were gathered, housed at an Irish research farm. A neck-mounted accelerometer was used to gather behavioral metrics, additional automatically recorded data consisted of milk production and live weight. Locomotion scoring data were manually recorded, using a one-to-five scale (1 = non-lame, 5 = severely lame). Locomotion scores where then used to label the cows as sound (locomotion score 1) or unsound (locomotion score ≥ 2). Four supervised classification models, using a gradient boosted decision tree machine learning algorithm, were constructed to investigate whether cows could be classified as sound or unsound. Data available for model building included behavioral metrics, milk production and animal characteristics.

RESULTS

The resulting models were constructed using various combinations of the data sources. The accuracy of the models was then compared using confusion matrices, receiver-operator characteristic curves and calibration plots. The model which achieved the highest performance according to the accuracy measures, was the model combining all the available data, resulting in an area under the curve of 85% and a sensitivity and specificity of 78%.

CONCLUSION

These results show that 85% of this model's predictions were correct in identifying cows as sound or unsound, showing that the use of a neck-mounted accelerometer, in combination with production and other animal data, has potential to replace visual locomotion scoring as lameness detection method in dairy cows.

Developing a Simulated Online Model That Integrates GNSS, Accelerometer and Weather Data to Detect Parturition Events in Grazing Sheep: A Machine Learning Approach

Simple Summary

Near-real-time monitoring of livestock using on-animal sensor technology has the potential to improve animal welfare and productivity through increased surveillance and improved decision-making capabilities. One potentially valuable application is for monitoring of lambing events in sheep. This research reports on the development of a machine learning classification algorithm for autonomous detection of lambing events. The algorithm uses data from Global Navigation Satellite System (GNSS) tracking collars, accelerometer ear tags and local weather data. Overall, four features of sheep behaviour were identified as having the greatest importance for lambing detection, including various measures of social distancing and frequency of posture change. Using these four features, the final algorithm was able to detect up to 91% of lambing events. This knowledge is intended to contribute to the development of commercially feasible lambing detection systems for improved surveillance of animals, ultimately improving methods of monitoring during critical welfare periods.

Abstract

In the current study, a simulated online parturition detection model is developed and reported. Using a machine learning (ML)-based approach, the model incorporates data from Global Navigation Satellite System (GNSS) tracking collars, accelerometer ear tags and local weather data, with the aim of detecting parturition events in pasture-based sheep. The specific objectives were two-fold: (i) determine which sensor systems and features provide the most useful information for lambing detection; (ii) evaluate how these data might be integrated using ML classification to alert to a parturition event as it occurs. Two independent field trials were conducted during the 2017 and 2018 lambing seasons in New Zealand, with the data from each used for ML training and independent validation, respectively. Based on objective (i), four features were identified as exerting the greatest importance for lambing detection: mean distance to peers (MDP), MDP compared to the flock mean (MDP.Mean), closest peer (CP) and posture change (PC). Using these four features, the final ML was able to detect 27% and 55% of lambing events within ±3 h of birth with no prior false positives. If the model sensitivity was manipulated such that earlier false positives were permissible, this detection increased to 91% and 82% depending on the requirement for a single alert, or two consecutive alerts occurring. To identify the potential causes of model failure, the data of three animals were investigated further. Lambing detection appeared to rely on increased social isolation behaviour in addition to increased PC behaviour. The results of the study support the use of integrated sensor data for ML-based detection of parturition events in grazing sheep. This is the first known application of ML classification for the detection of lambing in pasture-based sheep. Application of this knowledge could have significant impacts on the ability to remotely monitor animals in commercial situations, with a logical extension of the information for remote monitoring of animal welfare.

Machine Learning Algorithms to Classify and Quantify Multiple Behaviours in Dairy Calves Using a Sensor: Moving beyond Classification in Precision Livestock

Previous research has shown that sensors monitoring lying behaviours and feeding can detect early signs of ill health in calves. There is evidence to suggest that monitoring change in a single behaviour might not be enough for disease prediction. In calves, multiple behaviours such as locomotor play, self-grooming, feeding and activity whilst lying are likely to be informative. However, these behaviours can occur rarely in the real world, which means simply counting behaviours based on the prediction of a classifier can lead to overestimation. Here, we equipped thirteen pre-weaned dairy calves with collar-mounted sensors and monitored their behaviour with video cameras. Behavioural observations were recorded and merged with sensor signals. Features were calculated for 1-10-s windows and an AdaBoost ensemble learning algorithm implemented to classify behaviours. Finally, we developed an adjusted count quantification algorithm to predict the prevalence of locomotor play behaviour on a test dataset with low true prevalence (0.27%). Our algorithm identified locomotor play (99.73% accuracy), self-grooming (98.18% accuracy), ruminating (94.47% accuracy), non-nutritive suckling (94.96% accuracy), nutritive suckling (96.44% accuracy), active lying (90.38% accuracy) and non-active lying (90.38% accuracy). Our results detail recommended sampling frequencies, feature selection and window size. The quantification estimates of locomotor play behaviour were highly correlated with the true prevalence (0.97; p < 0.001) with a total overestimation of 18.97%. This study is the first to implement machine learning approaches for multi-class behaviour identification as well as behaviour quantification in calves. This has potential to contribute towards new insights to evaluate the health and welfare in calves by use of wearable sensors.

Relating Lying Behavior With Climate, Body Condition Score, and Milk Production in Dairy Cows

Attention on animal behavior and welfare has been increasing. Scientific knowledge about the effect of behavior and welfare on animals' production augmented and made clear the need of improving their living conditions. Among the variables to monitor in dairy cattle farming, lying time represents a signal for health and welfare status as well as for milk production. The aim of this study is to identify the relationship among the lying behavior of dairy cows and milk production, body condition score (BCS), weather variables, and the temperature–humidity index (THI) in the barn from a dairy farm located in Northern Italy. One-year data were collected on this farm with sensors that allowed monitoring of the environmental conditions in the barn and the activity of primiparous lactating cows. Principal components analysis (PCA), factor analysis (FA), generalized linear model select (GLMSelect), and logistic analysis (LA) were carried out to get the relationships among variables. Among the main results, it emerges that the effect of weather parameters is quite restrained, except for THI > 70, which negatively affects the lying time. In addition, the most productive cows are found to lie down more than the less productive ones, and the parameters of milk production, lying time, and BCS are found to be linked by a similar trend.

Association of prepartum lying time with nonesterified fatty acids and stillbirth in prepartum dairy heifers and cows

The objective of the present study was to assess the association of prepartum lying time (LT) and the coefficient of variation (CV) of LT within 7 d before calving with prepartum serum nonesterified fatty acid (NEFA) concentration at 7 ± 3 d prepartum (dpp) and stillbirth. Prepartum pregnant Holstein heifers and cows from 3 dairy herds were used (n = 1,044). Animals were housed in freestall barns using a prepartum pen 21 d before the expected calving date and were moved into a contiguous maternity pen at parturition. Monthly, cohorts of 20 to 36 animals (heifers and cows combined) were enrolled at each farm and electronic data loggers (IceQube, IceRobotics, Edinburgh, UK) were fitted to the hind leg of individual animals to assess their behavioral activity. Stillbirth was defined as a calf born dead or died during the first 24 h after parturition in dams with normal gestation length. The LT was recorded for the last 7 dpp to assess differences among dams with stillbirth versus those with a calf born alive. Mean LT within 7 d before blood NEFA collection was assessed to determine the association with prepartum serum NEFA at 7 ± 3 dpp. Blood samples for the assessment of serum NEFA concentration were collected from prepartum animals at 14 ± 3 and at 7 ± 3 dpp. Blood samples for total serum calcium concentration were collected from postpartum cows within 48 h after parturition to assess differences among cows with stillbirth versus those with a calf born alive. Data were analyzed using CORR, MIXED, or GLIMMIX procedures of SAS (SAS Institute Inc., Cary, NC). Dams experiencing dystocic births had a greater proportion of stillbirth, but herd, parity, and season did not have an effect. Dams with a stillborn calf had reduced LT and increased CV of LT within the last 7 dpp compared with dams with a calf born alive, regardless of parity. Multiparous cows with a stillborn calf had higher prepartum serum NEFA concentration compared with multiparous cows with a calf born alive, but this did not differ for first-calf heifers. Regardless of parity, the proportion of postpartum cows with hypocalcemia was higher for dams with a stillborn calf compared with those with a calf born alive. Regardless of parity, LT of prepartum dams was strongly correlated with the CV of LT (as LT increased, the CV decreased), and prepartum dams with a mean LT between 11 and 15 h/d had the lowest serum NEFA concentration compared with dams with LT of 8 to 10 or >16 h/d. Serum NEFA concentrations at 7 ± 3 dpp was positively correlated with the CV of LT within 7 d before blood sample. These results show that the dam's prepartum LT and its consistency over time are associated with prepartum serum NEFA and calf survival at calving.

Inferring an animal's environment through biologging: quantifying the environmental influence on animal movement

BACKGROUND

Animals respond to environmental variation by changing their movement in a multifaceted way. Recent advancements in biologging increasingly allow for detailed measurements of the multifaceted nature of movement, from descriptors of animal movement trajectories (e.g., using GPS) to descriptors of body part movements (e.g., using tri-axial accelerometers). Because this multivariate richness of movement data complicates inference on the environmental influence on animal movement, studies generally use simplified movement descriptors in statistical analyses. However, doing so limits the inference on the environmental influence on movement, as this requires that the multivariate richness of movement data can be fully considered in an analysis.

METHODS

We propose a data-driven analytic framework, based on existing methods, to quantify the environmental influence on animal movement that can accommodate the multifaceted nature of animal movement. Instead of fitting a simplified movement descriptor to a suite of environmental variables, our proposed framework centres on predicting an environmental variable from the full set of multivariate movement data. The measure of fit of this prediction is taken to be the metric that quantifies how much of the environmental variation relates to the multivariate variation in animal movement. We demonstrate the usefulness of this framework through a case study about the influence of grass availability and time since milking on cow movements using machine learning algorithms.

RESULTS

We show that on a one-hour timescale 37% of the variation in grass availability and 33% of time since milking influenced cow movements. Grass availability mostly influenced the cows' neck movement during grazing, while time since milking mostly influenced the movement through the landscape and the shared variation of accelerometer and GPS data (e.g., activity patterns). Furthermore, this framework proved to be insensitive to spurious correlations between environmental variables in quantifying the influence on animal movement.

CONCLUSIONS

Not only is our proposed framework well-suited to study the environmental influence on animal movement; we argue that it can also be applied in any field that uses multivariate biologging data, e.g., animal physiology, to study the relationships between animals and their environment.

SUPPLEMENTARY INFORMATION

Supplementary information accompanies this paper at 10.1186/s40462-020-00228-4.

Graduate Student Literature Review: Evaluating the appropriate use of wearable accelerometers in research to monitor lying behaviors of dairy cows

Until recently, animal behavior has been studied through close and extensive observation of individual animals and has relied on subjective assessments. Wearable technologies that allow the automation of dairy cow behavior recording currently dominate the precision dairy technology market. Wearable accelerometers provide new opportunities in animal ethology using quantitative measures of dairy cow behavior. Recent research developments indicate that quantitative measures of behavior may provide new objective on-farm measures to assist producers in predicting, diagnosing, and managing disease or injury on farms and allowing producers to monitor cow comfort and estrus behavior. These recent research developments and a large increase in the availability of wearable accelerometers have led to growing interest of both researchers and producers in this technology. This review aimed to summarize the studies that have validated lying behavior derived from accelerometers and to describe the factors that should be considered when using leg-attached accelerometers and neck-worn collars to describe lying behavior (e.g., lying time and lying bouts) in dairy cows for research purposes. Specifically, we describe accelerometer technology, including the instrument properties and methods for recording motion; the raw data output from accelerometers; and methods developed for the transformation of raw data into meaningful and interpretable information. We highlight differences in validation study outcomes for researchers to consider when developing their own experimental methodology for the use of accelerometers to record lying behaviors in dairy cows. Finally, we discuss several factors that may influence the data recorded by accelerometers and highlight gaps in the literature. We conclude that researchers using accelerometers to record lying behaviors in dairy cattle should (1) select an accelerometer device that, based on device attachment and sampling rate, is appropriate to record the behavior of interest; (2) account for cow-, farm-, and management-related factors that could affect the lying behaviors recorded; (3) determine the appropriate editing criteria for the accurate interpretation of their data; (4) support their chosen method of recording, editing, and interpreting the data by referencing an appropriately designed and accurate validation study published in the literature; and (5) report, in detail, their methodology to ensure others can decipher how the data were captured and understand potential limitations of their methodology. We recommend that standardized protocols be developed for collecting, analyzing, and reporting lying behavior data recorded using wearable accelerometers for dairy cattle.