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

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.

Accelerometer systems as tools for health and welfare assessment in cattle and pigs - A review

Welfare assessment has traditionally been performed by direct observation by humans, providing information at only selected points in time. Recently, this assessment method has been questioned, as 'Precision Livestock Farming' technologies may be able to deliver more valid, reliable and feasible real-time data at the individual level and serve as early monitoring systems for animal welfare. The aim of this paper is to describe how accelerometers can be used for welfare assessment based on the principles of the Welfare Quality assessment protocol. Algorithm development is based mainly on the detection of behavioural traits. So far, high accuracies have been found for movement and resting behaviours in cows and pigs, while algorithm development for feeding and drinking behaviours in pigs lag behind progress in cows where valid algorithms are already available. Welfare studies have used accelerometer technology to address the effects on behaviour of diet, daily cycle, enrichment, housing, social mixing, oestrus, lameness and disease. Additional aspects to consider before a decision is made upon its use in research and in practical applications include battery life and sensor location. While accelerometer systems for cows are already being used by farmers, application in pigs has mainly remained at the research level.

The Internet of Things enhancing animal welfare and farm operational efficiency

The growth in wirelessly enabled sensor network technologies has enabled the low cost deployment of sensor platforms with applications in a range of sectors and communities. In the agricultural domain such sensors have been the foundation for the creation of decision support tools that enhance farm operational efficiency. This Research Reflection illustrates how these advances are assisting dairy farmers to optimise performance and illustrates where emerging sensor technology can offer additional benefits. One of the early applications for sensor technology at an individual animal level was the accurate identification of cattle entering into heat (oestrus) to increase the rate of successful pregnancies and thus optimise milk yield per animal. This was achieved through the use of activity monitoring collars and leg tags. Additional information relating to the behaviour of the cattle, namely the time spent eating and ruminating, was subsequently derived from collars giving further insights of economic value into the wellbeing of the animal, thus an enhanced range of welfare related services have been provisioned. The integration of the information from neck-mounted collars with the compositional analysis data of milk measured at a robotic milking station facilitates the early diagnosis of specific illnesses such as mastitis. The combination of different data streams also serves to eliminate the generation of false alarms, improving the decision making capability. The principle of integrating more data streams from deployed on-farm systems, for example, with feed composition data measured at the point of delivery using instrumented feeding wagons, supports the optimisation of feeding strategies and identification of the most productive animals. Optimised feeding strategies reduce operational costs and minimise waste whilst ensuring high welfare standards. These IoT-inspired solutions, made possible through Internet-enabled cloud data exchange, have the potential to make a major impact within farming practices. This paper gives illustrative examples and considers where new sensor technology from the automotive industry may also have a role.

A new direction for differentiating animal activity based on measuring angular velocity about the yaw axis

The use of animal-attached data loggers to quantify animal movement has increased in popularity and application in recent years. High-resolution tri-axial acceleration and magnetometry measurements have been fundamental in elucidating fine-scale animal movements, providing information on posture, traveling speed, energy expenditure, and associated behavioral patterns. Heading is a key variable obtained from the tandem use of magnetometers and accelerometers, although few field investigations have explored fine-scale changes in heading to elucidate differences in animal activity (beyond the notable exceptions of dead-reckoning).This paper provides an overview of the value and use of animal heading and a prime derivative, angular velocity about the yaw axis, as an important element for assessing activity extent with potential to allude to behaviors, using "free-ranging" Loggerhead turtles (Caretta caretta) as a model species.We also demonstrate the value of yaw rotation for assessing activity extent, which varies over the time scales considered and show that various scales of body rotation, particularly rate of change of yaw, can help resolve differences between fine-scale behavior-specific movements. For example, oscillating yaw movements about a central point of the body's arc implies bouts of foraging, while unusual circling behavior, indicative of conspecific interactions, could be identified from complete revolutions of the longitudinal axis.We believe this approach should help identification of behaviors and "space-state" approaches to enhance our interpretation of behavior-based movements, particularly in scenarios where acceleration metrics have limited value, such as for slow-moving animals.

Behavioural inference from signal processing using animal-borne multi-sensor loggers: a novel solution to extend the knowledge of sea turtle ecology

The identification of sea turtle behaviours is a prerequisite to predicting the activities and time-budget of these animals in their natural habitat over the long term. However, this is hampered by a lack of reliable methods that enable the detection and monitoring of certain key behaviours such as feeding. This study proposes a combined approach that automatically identifies the different behaviours of free-ranging sea turtles through the use of animal-borne multi-sensor recorders (accelerometer, gyroscope and time-depth recorder), validated by animal-borne video-recorder data. We show here that the combination of supervised learning algorithms and multi-signal analysis tools can provide accurate inferences of the behaviours expressed, including feeding and scratching behaviours that are of crucial ecological interest for sea turtles. Our procedure uses multi-sensor miniaturized loggers that can be deployed on free-ranging animals with minimal disturbance. It provides an easily adaptable and replicable approach for the long-term automatic identification of the different activities and determination of time-budgets in sea turtles. This approach should also be applicable to a broad range of other species and could significantly contribute to the conservation of endangered species by providing detailed knowledge of key animal activities such as feeding, travelling and resting.

The new design of cows' behavior classifier based on acceleration data and proposed feature set

Monitor and classify behavioral activities in cows is a helpful support solution for livestock based on the analysis of data from sensors attached to the animal. Accelerometers are particularly suited for monitoring cow behaviors due to small size, lightweight and high accuracy. Nevertheless, the interpretation of the data collected by such sensors when characterizing the type of behaviors still brings major challenges to developers, related to activity complexity (i.e., certain behaviors contain similar gestures). This paper presents a new design of cows' behavior classifier based on acceleration data and proposed feature set. Analysis of cow acceleration data is used to extract features for classification using machine learning algorithms. We found that with 5 features (mean, standard deviation, root mean square, median, range) and 16-second window of data (1 sample/second), classification of seven cow behaviors (including feeding, lying, standing, lying down, standing up, normal walking, active walking) achieved the overall highest performance. We validated the results with acceleration data from a public source. Performance of our proposed classifier was evaluated and compared to existing ones in terms of the sensitivity, the accuracy, the positive predictive value, and the negative predictive value.

Identifying Sheep Activity from Tri-Axial Acceleration Signals Using a Moving Window Classification Model

Behaviour is a useful indicator of an individual animal’s overall wellbeing. There is widespread agreement that measuring and monitoring individual behaviour autonomously can provide valuable opportunities to trigger and refine on-farm management decisions. Conventionally, this has required visual observation of animals across a set time period. Technological advancements, such as animal-borne accelerometers, are offering 24/7 monitoring capability. Accelerometers have been used in research to quantify animal behaviours for a number of years. Now, technology and software developers, and more recently decision support platform providers, are integrating to offer commercial solutions for the extensive livestock industries. For these systems to function commercially, data must be captured, processed and analysed in sync with data acquisition. Practically, this requires a continuous stream of data or a duty cycled data segment and, from an analytics perspective, the application of moving window algorithms to derive the required classification. The aim of this study was to evaluate the application of a ‘clean state’ moving window behaviour state classification algorithm applied to 3, 5 and 10 second duration segments of data (including behaviour transitions), to categorise data emanating from collar, leg and ear mounted accelerometers on five Merino ewes. The model was successful at categorising grazing, standing, walking and lying behaviour classes with varying sensitivity, and no significant difference in model accuracy was observed between the three moving window lengths. The accuracy in identifying behaviour classes was highest for the ear-mounted sensor (86%–95%), followed by the collar-mounted sensor (67%–88%) and leg-mounted sensor (48%–94%). Between-sheep variations in classification accuracy confirm the sensor orientation is an important source of variation in all deployment modes. This research suggests a moving window classifier is capable of segregating continuous accelerometer signals into exclusive behaviour classes and may provide an appropriate data processing framework for commercial deployments.

Method for Classifying Behavior of Livestock on Fenced Temperate Rangeland in Northern China

Different livestock behaviors have distinct effects on grassland degradation. However, because direct observation of livestock behavior is time- and labor-intensive, an automated methodology to classify livestock behavior according to animal position and posture is necessary. We applied the Random Forest algorithm to predict livestock behaviors in the Horqin Sand Land by using Global Positioning System (GPS) and tri-axis accelerometer data and then confirmed the results through field observations. The overall accuracy of GPS models was 85% to 90% when the time interval was greater than 300–800 s, which was approximated to the tri-axis model (96%) and GPS-tri models (96%). In the GPS model, the linear backward or forward distance were the most important determinants of behavior classification, and nongrazing was less than 30% when livestock travelled more than 30–50 m over a 5-min interval. For the tri-axis accelerometer model, the anteroposterior acceleration (–3 m/s²) of neck movement was the most accurate determinant of livestock behavior classification. Using instantaneous acceleration of livestock body movement more precisely classified livestock behaviors than did GPS location-based distance metrics. When a tri-axis model is unavailable, GPS models will yield sufficiently reliable classification accuracy when an appropriate time interval is defined.

Making Sense of Pharmacovigilance and Drug Adverse Event Reporting: Comparative Similarity Association Analysis Using AI Machine Learning Algorithms in Dogs and Cats

Drug-associated adverse events cause approximately 30 billion dollars a year of added health care expense, along with negative health outcomes including patient death. This constitutes a major public health concern. The US Food and Drug Administration (FDA) requires drug labeling to include potential adverse effects for each newly developed drug product. With the advancement in incidence of adverse drug events (ADEs) and potential adverse drug events, published studies have mainly concluded potential ADEs from labeling documents obtained from the FDA's preapproval clinical trials, and very few analyzed their research work based on reported ADEs after widespread use of a drug to animal subjects. The aforesaid procedure of deriving practice based on information from preapproval labeling may misrepresent or deprecate the incidence and prevalence of specific ADEs. In this study, we make the most of the recently disseminated ADE data by the FDA for animal drugs and devices used in animals to address this public and welfare concern. For this purpose, we implemented 5 different methods (Pearson distance, Spearman distance, cosine distance, Yule distance, and Euclidean distance) to determine the most efficient and robust approach to properly discover highly associated ADEs from the reported data and accurately exclude noise-induced reported events, while maintaining a high level of correlation precision. Our comparative analysis of ADEs based on an artificial intelligence (AI) approach for the 5 robust similarity methods revealed high ADE associations for 2 drugs used in dogs and cats. In addition, the described distance methods systematically analyzed and compared ADEs from the drug labeling sections with a specific emphasis on analyzing serious ADEs. Our finding showed that the cosine method significantly outperformed all the other methods by correctly detecting and validating ADEs based on the comparative similarity association analysis compared with ADEs reported by preapproval clinical trials, premarket testing, or postapproval complication experience of FDA-approved animal drugs.

Horsing Around—A Dataset Comprising Horse Movement

Movement data were collected at a riding stable over seven days. The dataset comprises data from 18 individual horses and ponies with 1.2 million 2-s data samples, of which 93,303 samples have been tagged with labels (labeled data). Data from 11 subjects were labeled. The data from six subjects and six activities were labeled more extensively. Data were collected during horse riding sessions and when the horses freely roamed the pasture over seven days. Sensor devices were attached to a collar that was positioned around the neck of horses. The orientation of the sensor devices was not strictly fixed. The sensors devices contained a three-axis accelerometer, gyroscope, and magnetometer and were sampled at 100 Hz.

Behavioural compass: animal behaviour recognition using magnetometers

BACKGROUND

Animal-borne data loggers today often house several sensors recording simultaneously at high frequency. This offers opportunities to gain fine-scale insights into behaviour from individual-sensor as well as integrated multi-sensor data. In the context of behaviour recognition, even though accelerometers have been used extensively, magnetometers have recently been shown to detect specific behaviours that accelerometers miss. The prevalent constraint of limited training data necessitates the importance of identifying behaviours with high robustness to data from new individuals, and may require fusing data from both these sensors. However, no study yet has developed an end-to-end approach to recognise common animal behaviours such as foraging, locomotion, and resting from magnetometer data in a common classification framework capable of accommodating and comparing data from both sensors.

METHODS

We address this by first leveraging magnetometers' similarity to accelerometers to develop biomechanical descriptors of movement: we use the static component given by sensor tilt with respect to Earth's local magnetic field to estimate posture, and the dynamic component given by change in sensor tilt with time to characterise movement intensity and periodicity. We use these descriptors within an existing hybrid scheme that combines biomechanics and machine learning to recognise behaviour. We showcase the utility of our method on triaxial magnetometer data collected on ten wild Kalahari meerkats (Suricata suricatta), with annotated video recordings of each individual serving as groundtruth. Finally, we compare our results with accelerometer-based behaviour recognition.

RESULTS

The overall recognition accuracy of > 94% obtained with magnetometer data was found to be comparable to that achieved using accelerometer data. Interestingly, higher robustness to inter-individual variability in dynamic behaviour was achieved with the magnetometer, while the accelerometer was better at estimating posture.

CONCLUSIONS

Magnetometers were found to accurately identify common behaviours, and were particularly robust to dynamic behaviour recognition. The use of biomechanical considerations to summarise magnetometer data makes the hybrid scheme capable of accommodating data from either or both sensors within the same framework according to each sensor's strengths. This provides future studies with a method to assess the added benefit of using magnetometers for behaviour recognition.

Music Festival Makes Hedgehogs Move: How Individuals Cope Behaviorally in Response to Human-Induced Stressors

Simple Summary

Mega-events like concerts or festivals can still impact wildlife even when protective measures are taken. We remotely observed eight hedgehogs in a Berlin city park before and during a music festival using measuring devices attached to their bodies. While the actual festival only lasted two days (with about 70,000 visitors each day), setting the area up and removing the stages and stalls took 17 days in total. Construction work continued around the clock, causing an increase in light, noise and human presence throughout the night. In response, the hedgehogs showed clear changes in their behavior in comparison to the 19-day period just before the festival. We found, however, that different individuals responded differently to these changes in their environment. This individuality and behavioral flexibility could be one reason why hedgehogs are able to live in big cities.

Abstract

Understanding the impact of human activities on wildlife behavior and fitness can improve their sustainability. In a pilot study, we wanted to identify behavioral responses to anthropogenic stress in an urban species during a semi-experimental field study. We equipped eight urban hedgehogs (Erinaceus europaeus; four per sex) with bio-loggers to record their behavior before and during a mega music festival (2 × 19 days) in Treptower Park, Berlin. We used GPS (Global Positioning System) to monitor spatial behavior, VHF (Very High Frequency)-loggers to quantify daily nest utilization, and accelerometers to distinguish between different behaviors at a high resolution and to calculate daily disturbance (using Degrees of Functional Coupling). The hedgehogs showed clear behavioral differences between the pre-festival and festival phases. We found evidence supporting highly individual strategies, varying between spatial and temporal evasion of the disturbance. Averaging the responses of the individual animals or only examining one behavioral parameter masked these potentially different individual coping strategies. Using a meaningful combination of different minimal-invasive bio-logger types, we were able to show high inter-individual behavioral variance of urban hedgehogs in response to an anthropogenic disturbance, which might be a precondition to persist successfully in urban environments.

Modern Virtual Fencing Application: Monitoring and Controlling Behavior of Goats Using GPS Collars and Warning Signals

This paper describes our virtual fence system for goats. The present invention is a method of controlling goats without visible physical fences and monitoring their condition. Control occurs through affecting goats, using one or more sound signals and electric shocks when they attempt to enter a restricted zone. One of the best Machine Learning (ML) classifications named Support Vector Machines (SVM) is used to observe the condition. A virtual fence boundary can be of any geometrical shape. A smart collar on goats’ necks can be detected by using a virtual fence application. Each smart collar consists of a global positioning system (GPS), an XBee communication module, an mp3 player, and an electrical shocker. Stimuli and classification results are presented from on-farm experiments with a goat equipped with smart collar. Using the proposed stimuli methods, we showed that the probability of a goat receiving an electrical stimulus following an audio cue (dog and emergency sounds) was low (20%) and declined over the testing period. Besides, the RBF kernel-based SVM classification model classified lying behavior with an extremely high classification accuracy (F-score of 1), whilst grazing, running, walking, and standing behaviors were also classified with a high accuracy (F-score of 0.95, 0.97, 0.81, and 0.8, respectively).

Dairy cattle behavior classifications based on decision tree learning using 3-axis neck-mounted accelerometers

Demand has been increasing recently for an automated monitoring system of animal behavior as a tool for the management of livestock animals. This study investigated the association between the behavior of dairy cattle and the acceleration data collected using three-axis neck-mounted accelerometers, as well as the feasibility of improving the precision of behavior classifications through machine learning. In total 38 Holstein dairy cows were used, and kept in four different farms. A logger was mounted to each collar to obtain acceleration data for calculating the activity level and variations. At the same time the behavior of the cattle was observed visually. Characteristic acceleration waves were recorded for eating, rumination, and lying, respectively; and the activity level and variations were significantly different among these behaviors (p < 0.01). Decision tree learning was performed on the data set from Farm A and validated its precision; which proved to be 99.2% in cross-validation, and 100% in test data sets from Farms B to D. This study showed that highly precise classifications for eating, rumination, and lying is possible by using decision tree learning to calculate the activity level and variations of cattle based on the data obtained by three-axis accelerometers mounted to a collar.

Validity of Micro-Data Loggers to Determine Walking Activity of Turkeys and Effects on Turkey Gait

Accelerometers have the potential to provide objective, non-invasive methods for detecting changes in animal behavior and health. Our objectives were to: (1) determine the effects of micro-acceleration data loggers (accelerometers) and habituation to accelerometers on turkey gait and health status, (2) determine age-related changes in gait and health status, and (3) assess the validity and reliability of the accelerometers. Thirty-six male commercial turkeys were randomly assigned to one of five groups: accelerometer and habituation period (AH), accelerometer and no habituation period (AN), VetRap bandage (no accelerometer) and habituation period (VH), bandage (no accelerometer) and no habituation period (VN), and nothing on either leg (C). Health status and body condition were assessed prior to video-recording birds as they walked across a Tekscan® pressure pad at 8, 12, and 16 weeks to determine effects of treatment on number of steps, cadence, gait time, gait distance, gait velocity, impulse, gait cycle time, maximum force, peak vertical pressure, single support time, contact time, step length, step time, step velocity, stride length, total double support time, and duty factor. Accelerometer validity and reliability were determined by comparing the number of steps detected by the accelerometer to the number of steps determined from video recordings. Several age-related changes in turkey gait were found regardless of habituation including a slower cadence at 16 weeks, shorter gait distance at 8 weeks, and slower gait velocity at 16 weeks. When comparing bandaged vs. unbandaged limbs, both treatment and age-treatment interactions were found depending on the gait parameter. Accelerometer validity and reliability were affected by both age and treatment. False discovery rate increased, while accuracy and specificity decreased with age. Validity and reliability were lowest for non-habituated birds (AN and VN). Results demonstrated that micro-data loggers do not adversely affect turkey health status, but habituation to wearing accelerometers greatly affects accelerometer reliability and validity. Accelerometer validity and turkey gait are also greatly affected by the age of the turkeys.

Automatic lameness detection in cattle

There is an increasing demand for health and welfare monitoring in modern dairy farming. The development of various innovative techniques aims at improving animal behaviour monitoring and thus animal welfare indicators on-farm. Automated lameness detection systems have to be valid, reliable and practicable to be applied in veterinary practice or under farm conditions. The objective of this literature review was to describe the current automated systems for detection of lameness in cattle, which have been recently developed and investigated for application in dairy research and practice. The automatic methods of lameness detection broadly fall into three categories: kinematic, kinetic and indirect methods. The performance of the methods were compared with the reference standard (locomotion score and/or lesion score) and evaluated based on level-based scheme defining the degree of development (level I, sensor technique; level II, validation of algorithm; level III, performance for detection of lameness and/or lesion; level IV, decision support with early warning system). Many scientific studies have been performed on levels I-III, but there are no studies of level IV technology. The adoption rate of automated lameness detection systems by herd managers mainly yields returns on investment by the early identification of lame cows. Long-term studies, using validated automated lameness detection systems aiming at early lameness detection, are still needed in order to improve welfare and production under field conditions.

Application of accelerometers to record drinking behaviour of beef cattle

Accelerometers have been used to record many cattle postures and behaviours including standing, lying, walking, grazing and ruminating but not cattle drinking behaviour. This study explores whether neck-mounted triaxial accelerometers can identify drinking and whether head-neck position and activity can be used to record drinking. Over three consecutive days, data were collected from 12 yearling Brahman cattle each fitted with a collar containing an accelerometer. Each day the cattle were herded into a small yard containing a water trough and allowed 5 min to drink. Drinking, standing (head up), walking and standing (head down) were recorded. Examination of the accelerometer data showed that drinking events were characterised by a unique signature compared with the other behaviours. A linear mixed-effects model identified two variables that reflected differences in head-neck position and activity between drinking and the other behaviours: mean of the z- (front-to-back) axis and variance of the x- (vertical) axis (P &lt; 0.05). Threshold values, derived from Kernel density plots, were applied to classify drinking from the other behaviours using these two variables. The method accurately classified drinking from standing (head up) with 100% accuracy, from walking with 92% accuracy and from standing (head down) with 79% accuracy. The study shows that accelerometers have the potential to record cattle drinking behaviour. Further development of a classification method for drinking is required to allow accelerometer-derived data to be used to improve our understanding of cattle drinking behaviour and ensure that their water intake needs are met.

High accuracy at low frequency: detailed behavioural classification from accelerometer data

Accelerometers are a valuable tool for studying animal behaviour and physiology where direct observation is unfeasible. However, giving biological meaning to multivariate acceleration data is challenging. Here, we describe a method that reliably classifies a large number of behaviours using tri-axial accelerometer data collected at the low sampling frequency of 1 Hz, using the dingo (Canis dingo) as an example. We used out-of-sample validation to compare the predictive performance of four commonly used classification models (random forest, k-nearest neighbour, support vector machine, and naïve Bayes). We tested the importance of predictor variable selection and moving window size for the classification of each behaviour and overall model performance. Random forests produced the highest out-of-sample classification accuracy, with our best-performing model predicting 14 behaviours with a mean accuracy of 87%. We also investigated the relationship between overall dynamic body acceleration (ODBA) and the activity level of each behaviour, given the increasing use of ODBA in ecophysiology as a proxy for energy expenditure. ODBA values for our four 'high activity' behaviours were significantly greater than all other behaviours, with an overall positive trend between ODBA and intensity of movement. We show that a random forest model of relatively low complexity can mitigate some major challenges associated with establishing meaningful ecological conclusions from acceleration data. Our approach has broad applicability to free-ranging terrestrial quadrupeds of comparable size. Our use of a low sampling frequency shows potential for deploying accelerometers over extended time periods, enabling the capture of invaluable behavioural and physiological data across different ontogenies.

Smart Animal Agriculture: Application of Real-Time Sensors to Improve Animal Well-Being and Production

Consumption of animal products such as meat, milk, and eggs in first-world countries has leveled off, but it is rising precipitously in developing countries. Agriculture will have to increase its output to meet demand, opening the door to increased automation and technological innovation; intensified, sustainable farming; and precision livestock farming (PLF) applications. Early indicators of medical problems, which use sensors to alert cattle farmers early concerning individual animals that need special care, are proliferating. Wearable technologies dominate the market. In less-value-per-animal systems like sheep, goat, pig, poultry, and fish, one sensor, like a camera or robot per herd/flock/school, rather than one sensor per animal, will become common. PLF sensors generate huge amounts of data, and many actors benefit from PLF data. No standards currently exist for sharing sensor-generated data, limiting the use of commercial sensors. Technologies providing accurate data can enhance a well-managed farm. Development of methods to turn the data into actionable solutions is critical.

Feature Selection and Comparison of Machine Learning Algorithms in Classification of Grazing and Rumination Behaviour in Sheep

Grazing and ruminating are the most important behaviours for ruminants, as they spend most of their daily time budget performing these. Continuous surveillance of eating behaviour is an important means for monitoring ruminant health, productivity and welfare. However, surveillance performed by human operators is prone to human variance, time-consuming and costly, especially on animals kept at pasture or free-ranging. The use of sensors to automatically acquire data, and software to classify and identify behaviours, offers significant potential in addressing such issues. In this work, data collected from sheep by means of an accelerometer/gyroscope sensor attached to the ear and collar, sampled at 16 Hz, were used to develop classifiers for grazing and ruminating behaviour using various machine learning algorithms: random forest (RF), support vector machine (SVM), k nearest neighbour (kNN) and adaptive boosting (Adaboost). Multiple features extracted from the signals were ranked on their importance for classification. Several performance indicators were considered when comparing classifiers as a function of algorithm used, sensor localisation and number of used features. Random forest yielded the highest overall accuracies: 92% for collar and 91% for ear. Gyroscope-based features were shown to have the greatest relative importance for eating behaviours. The optimum number of feature characteristics to be incorporated into the model was 39, from both ear and collar data. The findings suggest that one can successfully classify eating behaviours in sheep with very high accuracy; this could be used to develop a device for automatic monitoring of feed intake in the sheep sector to monitor health and welfare.

Development and validation of an ensemble classifier for real-time recognition of cow behavior patterns from accelerometer data and location data

Behaviors are important indicators for assessing the health and well-being of dairy cows. The aim of this study is to develop and validate an ensemble classifier for automatically measuring and distinguishing several behavior patterns of dairy cows from accelerometer data and location data. The ensemble classifier consists of two parts, our new Multi-BP-AdaBoost algorithm and a data fusion method based on D-S evidence theory. We identify seven behavior patterns: feeding, lying, standing, lying down, standing up, normal walking, and active walking. Accuracy, sensitivity, and precision were used to validate classification performance. The Multi-BP-AdaBoost algorithm performed well when identifying lying (92% accuracy, 93% sensitivity, 82% precision), lying down (99%, 82%, 86%), standing up (99%, 74%, 85%), normal walking (97%, 92%, 86%), and active walking (99%, 94%, 89%). Its results were poor for feeding (80%, 52%, 55%) and standing (80%, 46%, 58%), which are difficult to differentiate using a leg-mounted sensor. Position data made it possible to differentiate feeding and standing. The D-S evidence fusion method for combining accelerometer data and location data in classification was used to fuse two pieces of basic behavior-related evidence into a single estimation model. With this addition, the sensitivity and precision of the two difficult behaviors increased by approximately 20 percentage points. In conclusion, the classification results indicate that the ensemble classifier effectively recognizes various behavior patterns in dairy cows. However, further work is needed to study the robustness of the feature and model by increasing the number of cows enrolled in the trial.

Multi-dimensional Precision Livestock Farming: a potential toolbox for sustainable rangeland management

BACKGROUND

Precision Livestock Farming (PLF) is a promising approach to minimize the conflicts between socio-economic activities and landscape conservation. However, its application on extensive systems of livestock production can be challenging. The main difficulties arise because animals graze on large natural pastures where they are exposed to competition with wild herbivores for heterogeneous and scarce resources, predation risk, adverse weather, and complex topography. Considering that the 91% of the world's surface devoted to livestock production is composed of extensive systems (i.e., rangelands), our general aim was to develop a PLF methodology that quantifies: (i) detailed behavioural patterns, (ii) feeding rate, and (iii) costs associated with different behaviours and landscape traits.

METHODS

For this, we used Merino sheep in Patagonian rangelands as a case study. We combined data from an animal-attached multi-sensor tag (tri-axial acceleration, tri-axial magnetometry, temperature sensor and Global Positioning System) with landscape layers from a Geographical Information System to acquire data. Then, we used high accuracy decision trees, dead reckoning methods and spatial data processing techniques to show how this combination of tools could be used to assess energy balance, predation risk and competition experienced by livestock through time and space.

RESULTS

The combination of methods proposed here are a useful tool to assess livestock behaviour and the different factors that influence extensive livestock production, such as topography, environmental temperature, predation risk and competition for heterogeneous resources. We were able to quantify feeding rate continuously through time and space with high accuracy and show how it could be used to estimate animal production and the intensity of grazing on the landscape. We also assessed the effects of resource heterogeneity (inferred through search times), and the potential costs associated with predation risk, competition, thermoregulation and movement on complex topography.

DISCUSSION

The quantification of feeding rate and behavioural costs provided by our approach could be used to estimate energy balance and to predict individual growth, survival and reproduction. Finally, we discussed how the information provided by this combination of methods can be used to develop wildlife-friendly strategies that also maximize animal welfare, quality and environmental sustainability.

Identification of Griffon Vulture's Flight Types Using High-Resolution Tracking Data

Being one of the most frequently killed raptors by collision with wind turbines, little is known about the Griffon vulture's flight strategies and behaviour in a fine scale. In this study, we used high-resolution tracking data to differentiate between the most frequently observed flight types of the Griffon, and evaluated the performance of our proposed approach by an independent observation during a period of 4 weeks of fieldwork. Five passive flight types including three types of soaring and two types of gliding were discriminated using the patterns of measured GPS locations. Of all flight patterns, gliding was classified precisely (precision = 88%), followed by linear and thermal soaring with precision of 83 and 75%, respectively. The overall accuracy of our classification was 70%. Our study contributes a baseline technique using high-resolution tracking data for the classification of flight types, and is one step forward towards the collision management of this species.

Support vector machines for explaining physiological stress response in Wood mice (Apodemus sylvaticus)

Physiological stress response is a crucial adaptive mechanism for prey species survival. This paper aims to identify the main environmental and/or individual factors better explaining the stress response in Wood mice, Apodemus sylvaticus. We analyzed alterations in fecal glucocorticoid metabolite (FCM) concentration - extensively used as an accurate measure of the physiological stress response - of wild mice fecal samples seasonally collected during three years. Then, support vector machines were built to predict said concentration according to different stressors. These statistical tools appear to be particularly suitable for small datasets with substantial number of dimensions, corroborating that the stress response is an extremely complex process in which multiple factors can simultaneously partake in a context-dependent manner, i.e., the role of each potential stressor varies in time depending on other stressors. However, air-humidity, temperature and body-weight allowed us to explain the FCM fluctuation in 98% of our samples. The relevance of air-humidity and temperature altering FCM level could be linked to the presence of an abundant vegetation cover and, therefore, to food availability and predation risk perception. Body-weight might be related to the stress produced by reproduction and other intraspecific relationships such as social dominance or territorial behavior.

Evaluation of sampling frequency, window size and sensor position for classification of sheep behaviour

Automated behavioural classification and identification through sensors has the potential to improve health and welfare of the animals. Position of a sensor, sampling frequency and window size of segmented signal data has a major impact on classification accuracy in activity recognition and energy needs for the sensor, yet, there are no studies in precision livestock farming that have evaluated the effect of all these factors simultaneously. The aim of this study was to evaluate the effects of position (ear and collar), sampling frequency (8, 16 and 32 Hz) of a triaxial accelerometer and gyroscope sensor and window size (3, 5 and 7 s) on the classification of important behaviours in sheep such as lying, standing and walking. Behaviours were classified using a random forest approach with 44 feature characteristics. The best performance for walking, standing and lying classification in sheep (accuracy 95%, F-score 91%-97%) was obtained using combination of 32 Hz, 7 s and 32 Hz, 5 s for both ear and collar sensors, although, results obtained with 16 Hz and 7 s window were comparable with accuracy of 91%-93% and F-score 88%-95%. Energy efficiency was best at a 7 s window. This suggests that sampling at 16 Hz with 7 s window will offer benefits in a real-time behavioural monitoring system for sheep due to reduced energy needs.

Predicting Lameness in Sheep Activity Using Tri-Axial Acceleration Signals

Simple Summary

Monitoring livestock farmed under extensive conditions is challenging and this is particularly difficult when observing animal behaviour at an individual level. Lameness is a disease symptom that has traditionally relied on visual inspection to detect those animals with an abnormal walking pattern. More recently, accelerometer sensors have been used in other livestock industries to detect lame animals. These devices are able to record changes in activity intensity, allowing us to differentiate between a grazing, walking, and resting animal. Using these on-animal sensors, grazing, standing, walking, and lame walking were accurately detected from an ear attached sensor. With further development, this classification algorithm could be linked with an automatic livestock monitoring system to provide real time information on individual health status, something that is practically not possible under current extensive livestock production systems.

Abstract

Lameness is a clinical symptom associated with a number of sheep diseases around the world, having adverse effects on weight gain, fertility, and lamb birth weight, and increasing the risk of secondary diseases. Current methods to identify lame animals rely on labour intensive visual inspection. The aim of this current study was to determine the ability of a collar, leg, and ear attached tri-axial accelerometer to discriminate between sound and lame gait movement in sheep. Data were separated into 10 s mutually exclusive behaviour epochs and subjected to Quadratic Discriminant Analysis (QDA). Initial analysis showed the high misclassification of lame grazing events with sound grazing and standing from all deployment modes. The final classification model, which included lame walking and all sound activity classes, yielded a prediction accuracy for lame locomotion of 82%, 35%, and 87% for the ear, collar, and leg deployments, respectively. Misclassification of sound walking with lame walking within the leg accelerometer dataset highlights the superiority of an ear mode of attachment for the classification of lame gait characteristics based on time series accelerometer data.

Development and application of a machine learning algorithm for classification of elasmobranch behaviour from accelerometry data

Discerning behaviours of free-ranging animals allows for quantification of their activity budget, providing important insight into ecology. Over recent years, accelerometers have been used to unveil the cryptic lives of animals. The increased ability of accelerometers to store large quantities of high resolution data has prompted a need for automated behavioural classification. We assessed the performance of several machine learning (ML) classifiers to discern five behaviours performed by accelerometer-equipped juvenile lemon sharks (Negaprion brevirostris) at Bimini, Bahamas (25°44'N, 79°16'W). The sharks were observed to exhibit chafing, burst swimming, headshaking, resting and swimming in a semi-captive environment and these observations were used to ground-truth data for ML training and testing. ML methods included logistic regression, an artificial neural network, two random forest models, a gradient boosting model and a voting ensemble (VE) model, which combined the predictions of all other (base) models to improve classifier performance. The macro-averaged F-measure, an indicator of classifier performance, showed that the VE model improved overall classification (F-measure 0.88) above the strongest base learner model, gradient boosting (0.86). To test whether the VE model provided biologically meaningful results when applied to accelerometer data obtained from wild sharks, we investigated headshaking behaviour, as a proxy for prey capture, in relation to the variables: time of day, tidal phase and season. All variables were significant in predicting prey capture, with predations most likely to occur during early evening and less frequently during the dry season and high tides. These findings support previous hypotheses from sporadic visual observations.

On the use of on-cow accelerometers for the classification of behaviours in dairy barns

Analysing behaviours can provide insight into the health and overall well-being of dairy cows. Automatic monitoring systems using e.g., accelerometers are becoming increasingly important to accurately quantify cows' behaviours as the herd size increases. The aim of this study is to automatically classify cows' behaviours by comparing leg- and neck-mounted accelerometers, and to study the effect of the sampling rate and the number of accelerometer axes logged on the classification performances. Lying, standing, and feeding behaviours of 16 different lactating dairy cows were logged for 6h with 3D-accelerometers. The behaviours were simultaneously recorded using visual observation and video recordings as a reference. Different features were extracted from the raw data and machine learning algorithms were used for the classification. The classification models using combined data of the neck- and the leg-mounted accelerometers have classified the three behaviours with high precision (80-99%) and sensitivity (87-99%). For the leg-mounted accelerometer, lying behaviour was classified with high precision (99%) and sensitivity (98%). Feeding was classified more accurately by the neck-mounted versus the leg-mounted accelerometer (precision 92% versus 80%; sensitivity 97% versus 88%). Standing was the most difficult behaviour to classify when only one accelerometer was used. In addition, the classification performances were not highly influenced when only X, X and Z, or Z and Y axes were used for the classification instead of three axes, especially for the neck-mounted accelerometer. Moreover, the accuracy of the models decreased with about 20% when the sampling rate was decreased from 1Hz to 0.05Hz.

Estimation of the energy expenditure of grazing ruminants by incorporating dynamic body acceleration into a conventional energy requirement system

The estimation of energy expenditure (EE) of grazing animals is of great importance for efficient animal management on pasture. In the present study, a method is proposed to estimate EE in grazing animals based on measurements of body acceleration of animals in combination with the conventional Agricultural and Food Research Council (AFRC) energy requirement system. Three-dimensional body acceleration and heart rate were recorded for tested animals under both grazing and housing management. An acceleration index, vectorial dynamic body acceleration (VeDBA), was used to calculate activity allowance (AC) during grazing and then incorporate it into the AFRC system to estimate the EE (EE derived from VeDBA [EE]) of the grazing animals. The method was applied to 3 farm ruminant species (7 cattle, 6 goats, and 4 sheep). Energy expenditure based on heart rate (EE) was also estimated as a reference. The result showed that larger VeDBA and heart rate values were obtained under grazing management, resulting in greater EE and EE under grazing management than under housing management. There were large differences between the EE estimated from the 2 methods, where EE values were greater than EE (averages of 163.4 and 142.5% for housing and grazing management, respectively); the EE was lower than the EE, whereas the increase in EE under grazing in comparison with housing conditions was larger than that in EE. These differences may have been due to the use of an equation for estimating EE derived under laboratory conditions and due to the presence of the effects of physiological, psychological, and environmental factors in addition to physical activity being included in measurements for the heart rate method. The present method allowed us to separate activity-specific EE (i.e., AC) from overall EE, and, in fact, AC under grazing management were about twice times as large as those under housing management for farm ruminant animals. There is evidence that the conventional energy system can predict fasting metabolism and the AC of housed animals based on accumulated research on energy metabolism and that VeDBA can quantify physical activity separately from other factors in animals on pasture. Therefore, the use of the VeDBA appears to be a precise way to predict activity-specific EE under grazing conditions, and the method incorporating acceleration index data with a conventional energy system can be a simple and useful method for estimation of EE in farm ruminants on pastures.

Seeing It All: Evaluating Supervised Machine Learning Methods for the Classification of Diverse Otariid Behaviours

Constructing activity budgets for marine animals when they are at sea and cannot be directly observed is challenging, but recent advances in bio-logging technology offer solutions to this problem. Accelerometers can potentially identify a wide range of behaviours for animals based on unique patterns of acceleration. However, when analysing data derived from accelerometers, there are many statistical techniques available which when applied to different data sets produce different classification accuracies. We investigated a selection of supervised machine learning methods for interpreting behavioural data from captive otariids (fur seals and sea lions). We conducted controlled experiments with 12 seals, where their behaviours were filmed while they were wearing 3-axis accelerometers. From video we identified 26 behaviours that could be grouped into one of four categories (foraging, resting, travelling and grooming) representing key behaviour states for wild seals. We used data from 10 seals to train four predictive classification models: stochastic gradient boosting (GBM), random forests, support vector machine using four different kernels and a baseline model: penalised logistic regression. We then took the best parameters from each model and cross-validated the results on the two seals unseen so far. We also investigated the influence of feature statistics (describing some characteristic of the seal), testing the models both with and without these. Cross-validation accuracies were lower than training accuracy, but the SVM with a polynomial kernel was still able to classify seal behaviour with high accuracy (>70%). Adding feature statistics improved accuracies across all models tested. Most categories of behaviour -resting, grooming and feeding—were all predicted with reasonable accuracy (52–81%) by the SVM while travelling was poorly categorised (31–41%). These results show that model selection is important when classifying behaviour and that by using animal characteristics we can strengthen the overall accuracy.

The private life of echidnas: using accelerometry and GPS to examine field biomechanics and assess the ecological impact of a widespread, semi-fossorial monotreme

The short-beaked echidna (Tachyglossus aculeatus) is a monotreme and therefore provides a unique combination of phylogenetic history, morphological differentiation and ecological specialisation for a mammal. The echidna has a unique appendicular skeleton, a highly specialised myrmecophagous lifestyle and a mode of locomotion that is neither typically mammalian nor reptilian, but has aspects of both lineages. We therefore were interested in the interactions of locomotor biomechanics, ecology and movements for wild, free-living short-beaked echidnas. To assess locomotion in its complex natural environment, we attached both GPS and accelerometer loggers to the back of echidnas in both spring and summer. We found that the locomotor biomechanics of echidnas is unique, with lower stride length and stride frequency than reported for similar-sized mammals. Speed modulation is primarily accomplished through changes in stride frequency, with a mean of 1.39 Hz and a maximum of 2.31 Hz. Daily activity period was linked to ambient air temperature, which restricted daytime activity during the hotter summer months. Echidnas had longer activity periods and longer digging bouts in spring compared with summer. In summer, echidnas had higher walking speeds than in spring, perhaps because of the shorter time suitable for activity. Echidnas spent, on average, 12% of their time digging, which indicates their potential to excavate up to 204 m3 of soil a year. This information highlights the important contribution towards ecosystem health, via bioturbation, of this widespread Australian monotreme.

Using accelerometers to remotely and automatically characterize behavior in small animals

Activity budgets in wild animals are challenging to measure via direct observation because data collection is time consuming and observer effects are potentially confounding. Although tri-axial accelerometers are increasingly employed for this purpose, their application in small-bodied animals has been limited by weight restrictions. Additionally, accelerometers engender novel complications, as a system is needed to reliably map acceleration to behaviors. In this study, we describe newly developed, tiny acceleration-logging devices (1.5-2.5 g) and use them to characterize behavior in two chipmunk species. We collected paired accelerometer readings and behavioral observations from captive individuals. We then employed techniques from machine learning to develop an automatic system for coding accelerometer readings into behavioral categories. Finally, we deployed and recovered accelerometers from free-living, wild chipmunks. This is the first time to our knowledge that accelerometers have been used to generate behavioral data for small-bodied (<100 g), free-living mammals.

Localizing Tortoise Nests by Neural Networks

The goal of this research is to recognize the nest digging activity of tortoises using a device mounted atop the tortoise carapace. The device classifies tortoise movements in order to discriminate between nest digging, and non-digging activity (specifically walking and eating). Accelerometer data was collected from devices attached to the carapace of a number of tortoises during their two-month nesting period. Our system uses an accelerometer and an activity recognition system (ARS) which is modularly structured using an artificial neural network and an output filter. For the purpose of experiment and comparison, and with the aim of minimizing the computational cost, the artificial neural network has been modelled according to three different architectures based on the input delay neural network (IDNN). We show that the ARS can achieve very high accuracy on segments of data sequences, with an extremely small neural network that can be embedded in programmable low power devices. Given that digging is typically a long activity (up to two hours), the application of ARS on data segments can be repeated over time to set up a reliable and efficient system, called Tortoise@, for digging activity recognition.