Surveying Wild Animals from Satellites, Manned Aircraft and Unmanned Aerial Systems (UASs): A Review

Using convolutional neural networks to count parrot nest-entrances on photographs from the largest known colony of Psittaciformes

Counting animal populations is fundamental to understand ecological processes. Counts make it possible to estimate the size of an animal population at specific points in time, which is essential information for understanding demographic change. However, in the case of large populations, counts are time-consuming, particularly if carried out manually. Here, we took advantage of convolutional neural networks (CNN) to count the total number of nest-entrances in 222 photographs covering the largest known Psittaciformes (Aves) colony in the world. We conducted our study at the largest Burrowing Parrot Cyanoliseus patagonus colony, located on a cliff facing the Atlantic Ocean in the vicinity of El Cóndor village, in north-eastern Patagonia, Argentina. We also aimed to investigate the distribution of nest-entrances along the cliff with the colony. For this, we used three CNN architectures, U-Net, ResUnet, and DeepLabv3. The U-Net architecture showed the best performance, counting a mean of 59,842 Burrowing Parrot nest-entrances across the colony, with a mean absolute error of 2.7 nest-entrances over the testing patches, measured as the difference between actual and predicted counts per patch. Compared to a previous study conducted at El Cóndor colony more than 20 years ago, the CNN architectures also detected noteworthy differences in the distribution of the nest-entrances along the cliff. We show that the strong changes observed in the distribution of nest-entrances are a measurable effect of a long record of human-induced disturbance to the Burrowing Parrot colony at El Cóndor. Given the paramount importance of the Burrowing Parrot colony at El Cóndor, which concentrates 71% of the world's population of this species, we advocate that it is imperative to reduce such a degree of disturbance before the parrots reach the limit of their capacity of adaptation.

Improved Re-Parameterized Convolution for Wildlife Detection in Neighboring Regions of Southwest China

To autonomously detect wildlife images captured by camera traps on a platform with limited resources and address challenges such as filtering out photos without optimal objects, as well as classifying and localizing species in photos with objects, we introduce a specialized wildlife object detector tailored for camera traps. This detector is developed using a dataset acquired by the Saola Working Group (SWG) through camera traps deployed in Vietnam and Laos. Utilizing the YOLOv6-N object detection algorithm as its foundation, the detector is enhanced by a tailored optimizer for improved model performance. We deliberately introduce asymmetric convolutional branches to enhance the feature characterization capability of the Backbone network. Additionally, we streamline the Neck and use CIoU loss to improve detection performance. For quantitative deployment, we refine the RepOptimizer to train a pure VGG-style network. Experimental results demonstrate that our proposed method empowers the model to achieve an 88.3% detection accuracy on the wildlife dataset in this paper. This accuracy is 3.1% higher than YOLOv6-N, and surpasses YOLOv7-T and YOLOv8-N by 5.5% and 2.8%, respectively. The model consistently maintains its detection performance even after quantization to the INT8 precision, achieving an inference speed of only 6.15 ms for a single image on the NVIDIA Jetson Xavier NX device. The improvements we introduce excel in tasks related to wildlife image recognition and object localization captured by camera traps, providing practical solutions to enhance wildlife monitoring and facilitate efficient data acquisition. Our current work represents a significant stride toward a fully automated animal observation system in real-time in-field applications.

Do you get what you see? Insights of using mAP to select architectures of pretrained neural networks for automated aerial animal detection

The vast amount of images generated by aerial imagery in the context of regular wildlife surveys nowadays require automatic processing tools. At the top of the mountain of different methods to automatically detect objects in images reigns deep learning's object detection. The recent focus given to this task has led to an influx of many different architectures of neural networks that are benchmarked against standard datasets like Microsoft's Common Objects in COntext (COCO). Performance on COCO, a large dataset of computer vision images, is given in terms of mean Average Precision (mAP). In this study, we use six pretrained networks to detect red deer from aerial images, three of which have never been used, to our knowledge, in a context of aerial wildlife surveys. We compare their performance along COCO's mAP and a common test metric in animal surveys, the F1-score. We also evaluate how dataset imbalance and background uniformity, two common difficulties in wildlife surveys, impact the performance of our models. Our results show that the mAP is not a reliable metric to select the best model to count animals in aerial images and that a counting-focused metric like the F1-score should be favored instead. Our best overall performance was achieved with Generalized Focal Loss (GFL). It scored the highest along both metrics, combining most accurate counting and localization (with average F1-score of 0.96 and 0.97 and average mAP scores of 0.77 and 0.89 on both datasets respectively) and is therefore very promising for future applications. While both imbalance and background uniformity improved the performance of our models, their combined effect had twice as much impact as the choice of architecture. This finding seems to confirm that the recent data-centric shift in the deep learning field could also lead to performance gains in wildlife surveys.

Review of harmful algal bloom effects on birds with implications for avian wildlife in the Chesapeake Bay region

The Chesapeake Bay, along the mid-Atlantic coast of North America, is the largest estuary in the United States and provides critical habitat for wildlife. In contrast to point and non-point source release of pesticides, metals, and industrial, personal care and household use chemicals on biota in this watershed, there has only been scant attention to potential exposure and effects of algal toxins on wildlife in the Chesapeake Bay region. As background, we first review the scientific literature on algal toxins and harmful algal bloom (HAB) events in various regions of the world that principally affected birds, and to a lesser degree other wildlife. To examine the situation for the Chesapeake, we compiled information from government reports and databases summarizing wildlife mortality events for 2000 through 2020 that were associated with potentially toxic algae and HAB events. Summary findings indicate that there have been few wildlife mortality incidents definitively linked to HABs, other mortality events that were suspected to be related to HABs, and more instances in which HABs may have indirectly contributed to or occurred coincident with wildlife mortality. The dominant toxins found in the Chesapeake Bay drainage that could potentially affect wildlife are microcystins, with concentrations in water approaching or exceeding human-based thresholds for ceasing recreational use and drinking water at a number of locations. As an increasing trend in HAB events in the U.S. and in the Chesapeake Bay have been reported, additional information on HAB toxin exposure routes, comparative sensitivity among species, consequences of sublethal exposure, and better diagnostic and risk criteria would greatly assist in predicting algal toxin hazard and risks to wildlife.

Computing in the Sky: A Survey on Intelligent Ubiquitous Computing for UAV-Assisted 6G Networks and Industry 4.0/5.0

Unmanned Aerial Vehicles (UAVs) are increasingly being used in a high-computation paradigm enabled with smart applications in the Beyond Fifth Generation (B5G) wireless communication networks. These networks have an avenue for generating a considerable amount of heterogeneous data by the expanding number of Internet of Things (IoT) devices in smart environments. However, storing and processing massive data with limited computational capability and energy availability at local nodes in the IoT network has been a significant difficulty, mainly when deploying Artificial Intelligence (AI) techniques to extract discriminatory information from the massive amount of data for different tasks.Therefore, Mobile Edge Computing (MEC) has evolved as a promising computing paradigm leveraged with efficient technology to improve the quality of services of edge devices and network performance better than cloud computing networks, addressing challenging problems of latency and computation-intensive offloading in a UAV-assisted framework. This paper provides a comprehensive review of intelligent UAV computing technology to enable 6G networks over smart environments. We highlight the utility of UAV computing and the critical role of Federated Learning (FL) in meeting the challenges related to energy, security, task offloading, and latency of IoT data in smart environments. We present the reader with an insight into UAV computing, advantages, applications, and challenges that can provide helpful guidance for future research.

Vision-Based Module for Herding with a Sheepdog Robot

Livestock farming is assisted more and more by technological solutions, such as robots. One of the main problems for shepherds is the control and care of livestock in areas difficult to access where grazing animals are attacked by predators such as the Iberian wolf in the northwest of the Iberian Peninsula. In this paper, we propose a system to automatically generate benchmarks of animal images of different species from iNaturalist API, which is coupled with a vision-based module that allows us to automatically detect predators and distinguish them from other animals. We tested multiple existing object detection models to determine the best one in terms of efficiency and speed, as it is conceived for real-time environments. YOLOv5m achieves the best performance as it can process 64 FPS, achieving an mAP (with IoU of 50%) of 99.49% for a dataset where wolves (predator) or dogs (prey) have to be detected and distinguished. This result meets the requirements of pasture-based livestock farms.

Eyes on the herd: Quantifying ungulate density from satellite, unmanned aerial systems, and GPScollar data

Novel approaches to quantifying density and distributions could help biologists adaptively manage wildlife populations, particularly if methods are accurate, consistent, cost-effective, rapid, and sensitive to change. Such approaches may also improve research on interactions between density and processes of interest, such as disease transmission across multiple populations. We assess how satellite imagery, unmanned aerial system (UAS) imagery, and Global Positioning System (GPS) collar data vary in characterizing elk density, distribution, and count patterns across times with and without supplemental feeding at the National Elk Refuge (NER) in the US state of Wyoming. We also present the first comparison of satellite imagery data with traditional counts for ungulates in a temperate system. We further evaluate seven different aggregation metrics to identify the most consistent and sensitive metrics for comparing density and distribution across time and populations. All three data sources detected higher densities and aggregation locations of elk during supplemental feeding than non-feeding at the NER. Kernel density estimates (KDEs), KDE polygon areas, and the first quantile of interelk distances detected differences with the highest sensitivity and were most highly correlated across data sources. Both UAS and satellite imagery provide snapshots of density and distribution patterns of most animals in the area at lower cost than GPS collars. While satellite-based counts were lower than traditional counts, aggregation metrics matched those from UAS and GPS data sources when animals appeared in high contrast to the landscape, including brown elk against new snow in open areas. UAS counts of elk were similar to traditional ground-based counts on feed grounds and are the best data source for assessing changes in small spatial extents. Satellite, UAS, or GPS data can provide appropriate data for assessing density and changes in density from adaptive management actions. For the NER, where high elk densities are beneath controlled airspace, GPS collar data will be most useful for evaluating how management actions, including changes in the dates of supplemental feeding, influence elk density and aggregation across large spatial extents. Using consistent and sensitive measures of density may improve research on the drivers and effects of density within and across a wide range of species.

Remote sensing techniques for automated marine mammals detection: a review of methods and current challenges

Marine mammals are under pressure from multiple threats, such as global climate change, bycatch, and vessel collisions. In this context, more frequent and spatially extensive surveys for abundance and distribution studies are necessary to inform conservation efforts. Marine mammal surveys have been performed visually from land, ships, and aircraft. These methods can be costly, logistically challenging in remote locations, dangerous to researchers, and disturbing to the animals. The growing use of imagery from satellite and unoccupied aerial systems (UAS) can help address some of these challenges, complementing crewed surveys and allowing for more frequent and evenly distributed surveys, especially for remote locations. However, manual counts in satellite and UAS imagery remain time and labor intensive, but the automation of image analyses offers promising solutions. Here, we reviewed the literature for automated methods applied to detect marine mammals in satellite and UAS imagery. The performance of studies is quantitatively compared with metrics that evaluate false positives and false negatives from automated detection against manual counts of animals, which allows for a better assessment of the impact of miscounts in conservation contexts. In general, methods that relied solely on statistical differences in the spectral responses of animals and their surroundings performed worse than studies that used convolutional neural networks (CNN). Despite mixed results, CNN showed promise, and its use and evaluation should continue. Overall, while automation can reduce time and labor, more research is needed to improve the accuracy of automated counts. With the current state of knowledge, it is best to use semi-automated approaches that involve user revision of the output. These approaches currently enable the best tradeoff between time effort and detection accuracy. Based on our analysis, we identified thermal infrared UAS imagery as a future research avenue for marine mammal detection and also recommend the further exploration of object-based image analysis (OBIA). Our analysis also showed that past studies have focused on the automated detection of baleen whales and pinnipeds and that there is a gap in studies looking at toothed whales, polar bears, sirenians, and mustelids.

Unmanned aerial vehicle surveys reveal unexpectedly high density of a threatened deer in a plantation forestry landscape

The Vulnerable marsh deer Blastocerus dichotomus, the largest native cervid in South America, is declining throughout its range as a result of the conversion of wetlands and overhunting. Estimated densities in open wetlands of several types are 0.1–6.8 individuals per km2. We undertook the first unmanned aerial vehicle (UAV) survey of the marsh deer to estimate the density of this species in a 113.6 km2 area under forestry management in the lower delta of the Paraná River, Argentina. During 6–8 August 2019, at a time of year when canopy cover is minimal, we surveyed marsh deer using Phantom 4 Pro UAVs along 94 transects totalling 127.8 km and 8.6 km2 (8.1% of the study area). The 5,506 photographs obtained were manually checked by us and by a group of 39 trained volunteers, following a standardized protocol. We detected a total of 58 marsh deer, giving an estimated density of 6.90 individuals per km2 (95% CI 5.26–8.54), which extrapolates to 559–908 individuals in our 113.6 km2 study area. As it has generally been assumed that marsh deer prefer open habitats, this relatively high estimate of density within a forestry plantation matrix is unexpected. We discuss the advantages of using UAVs to survey marsh deer and other related ungulates.

Foreword to the Special Issue on ‘The rapidly expanding role of drones as a tool for wildlife research’

Drones have emerged as a popular wildlife research tool, but their use for many species and environments remains untested and research is needed on validation of sampling approaches that are optimised for unpiloted aircraft. Here, we present a foreword to a special issue that features studies pushing the taxonomic and innovation boundaries of drone research and thus helps address these knowledge and application gaps. We then conclude by highlighting future drone research ideas that are likely to push biology and conservation in exciting new directions.

UAV-Enabled Mobile Edge-Computing for IoT Based on AI: A Comprehensive Review

Unmanned aerial vehicles (UAVs) are becoming integrated into a wide range of modern IoT applications. The growing number of networked IoT devices generates a large amount of data. However, processing and memorizing this massive volume of data at local nodes have been deemed critical challenges, especially when using artificial intelligence (AI) systems to extract and exploit valuable information. In this context, mobile edge computing (MEC) has emerged as a way to bring cloud computing (CC) processes within reach of users, to address computation-intensive offloading and latency issues. This paper provides a comprehensive review of the most relevant research works related to UAV technology applications in terms of enabled or assisted MEC architectures. It details the utility of UAV-enabled MEC architecture regarding emerging IoT applications and the role of both deep learning (DL) and machine learning (ML) in meeting various limitations related to latency, task offloading, energy demand, and security. Furthermore, throughout this article, the reader gains an insight into the future of UAV-enabled MEC, the advantages and the critical challenges to be tackled when using AI.

Roadkills as a Method to Monitor Raccoon Dog Populations

The raccoon dog (Nyctereutes procyonoides) is one of the most frequently killed species on Lithuanian roads. As an invasive species, up-to-date knowledge of population size, trends and spatial distribution is critically important both for species assessment and for the planning of control measures. In Lithuania, however, raccoon dog surveys have not been carried out since 1997. We investigated, therefore, whether roadkill counts on predefined routes could be used as a proxy for a survey. Our dataset includes survey numbers for the period 1956-1997, hunting bag sizes for 1965-2020 (including the spatial distribution of the hunting bag in 2018-2020) and roadkill data relating to 1551 individuals between 2002-2020. At the most local scale, that of the hunting areas of hunting clubs, correlations between the numbers of hunted and roadkilled individuals were negative and insignificant or absent. At the country scale, however, we found significant correlation both between the numbers surveyed and hunted in 1965-1997 (r = 0.88), and between those hunted and the number of roadkills in 2002-2020 (r = 0.56-0.69). Therefore, we consider that roadkill counts on predefined and stable routes may be used as a proxy for a survey at the country scale. Practical implementation of the method is proposed.

Application of a Drone Magnetometer System to Military Mine Detection in the Demilitarized Zone

We propose a magnetometer system fitted on an unmanned aerial vehicle (UAV, or drone) and a data-processing method for detecting metal antipersonnel landmines (M16) in the demilitarized zone (DMZ) in Korea, which is an undeveloped natural environment. The performance of the laser altimeter was improved so that the drone could fly at a low and stable altitude, even in a natural environment with dust and bushes, and a magnetometer was installed on a pendulum to minimize the effects of magnetic noise and vibration from the drone. At a flight altitude of 1 m, the criterion for M16 is 5 nT. Simple low-pass filtering eliminates magnetic swing noise due to pendulum motion, and the moving average method eliminates changes related to the heading of the magnetometer. Magnetic exploration was conducted in an actual mine-removal area near the DMZ in Korea, with nine magnetic anomalies of more than 5 nT detected and a variety of metallic substances found within a 1-m radius of each detection site. The proposed UAV-based landmine detection system is expected to reduce risk to detection personnel and shorten the landmine-detection period by providing accurate scientific information about the detection area prior to military landmine-detection efforts.

A Review of Unmanned Aerial Vehicle Low-Altitude Remote Sensing (UAV-LARS) Use in Agricultural Monitoring in China

Precision agriculture relies on the rapid acquisition and analysis of agricultural information. An emerging method of agricultural monitoring is unmanned aerial vehicle low-altitude remote sensing (UAV-LARS), which possesses significant advantages of simple construction, strong mobility, and high spatial-temporal resolution with synchronously obtained image and spatial information. UAV-LARS could provide a high degree of overlap between X and Y during key crop growth periods that is currently lacking in satellite and remote sensing data. Simultaneously, UAV-LARS overcomes the limitations such as small scope of ground platform monitoring. Overall, UAV-LARS has demonstrated great potential as a tool for monitoring agriculture at fine- and regional-scales. Here, we systematically summarize the history and current application of UAV-LARS in Chinese agriculture. Specifically, we outline the technical characteristics and sensor payload of the available types of unmanned aerial vehicles and discuss their advantages and limitations. Finally, we provide suggestions for overcoming current limitations of UAV-LARS and directions for future work.

Estimating animal population size with very high-resolution satellite imagery

Very high-resolution (VHR) satellite sensors can be used to estimate the size of animal populations, a critical factor in wildlife management, and acquire animal spatial distributions in an economical, easy, and precise way. We developed a method for satellite population size estimation that includes a noninvasive photogrammetry, from which the animal's average orthographic area is calculated, and an imagery interpretation method that estimates population size based on the ratio of an observed animal population area to the average individual area. As a proof of concept, we used this method to estimate the population size of Whooper Swans (Cygnus cygnus) in a national nature reserve in China. To reduce errors, the reserve was subdivided into regions of interest based on locations of Whooper Swan and background brightness. Estimates from the satellite pixels were compared with manual counts made over 2 years, at 3 locations, and in 3 land-cover types. Our results showed 1124 Whooper Swans occupied a national nature reserve on 15 February 2013, and the average percent error was 3.16% (SE = 1.37). These results demonstrate that our method produced robust data for population size estimation that were indistinguishable from manual count data. Our method may be used generally to estimate population sizes of visible and gregarious animals that exhibit high contrast relative to their environments and may inform estimations of populations in complex backgrounds.

Detecting and Tracking the Positions of Wild Ungulates Using Sound Recordings

Monitoring wild ungulates such as deer is a highly challenging issue faced by wildlife managers. Wild ungulates are increasing in number worldwide, causing damage to ecosystems. For effective management, the precise estimation of their population size and habitat is essential. Conventional methods used to estimate the population density of wild ungulates, such as the light census survey, are time-consuming with low accuracy and difficult to implement in harsh environments like muddy wetlands. On the other hand, unmanned aerial vehicles are difficult to use in areas with dense tree cover. Although the passive acoustic monitoring of animal sounds is commonly used to evaluate their diversity, the potential for detecting animal positions from their sound has not been sufficiently investigated. This study introduces a new technique for detecting and tracking deer position in the wild using sound recordings. The technique relies on the time lag among three recorders to estimate the position. A sound recording system was also developed to overcome the time drift problem in the internal clock of recorders, by receiving time information from GPS satellites. Determining deer position enables the elimination of repetitive calls from the same deer, thus providing a promising tool to track deer movement. The validation results revealed that the proposed technique can provide reasonable accuracy for the experimental and natural environment. The identification of deer calls in Oze National Park over a period of two hours emphasizes the great potential of the proposed technique to detect repetitive deer calls, and track deer movement. Hence, the technique is the first step toward designing an automated system for estimating the population of deer or other vocal animals using sound recordings.

Remotely Piloted Aircraft System (RPAS)-Based Wildlife Detection: A Review and Case Studies in Maritime Antarctica

Simple Summary

Remotely piloted aircraft systems (RPAS) have been successfully applied in wildlife monitoring with imaging sensors to improve or to supplement conventional field observations. To effectively utilize this technique, we reviewed previous studies related to wildlife detection with RPAS. First, this study provides an overview of the applications of RPAS for wild animal studies from the perspective of individual detection and population surveys as well as behavioral studies. In terms of the RPAS payload, applying thermal-imaging sensors was determined to be advantageous in detecting homeothermic animals due to the thermal contrast with background habitat using case studies detecting southern elephant seal (Mirounga leonina) using RGB and thermal imaging sensors in King George Island, maritime Antarctica.

Abstract

In wildlife biology, it is important to conduct efficient observations and quantitative monitoring of wild animals. Conventional wildlife monitoring mainly relies on direct field observations by the naked eyes or through binoculars, on-site image acquisition at fixed spots, and sampling or capturing under severe areal constraints. Recently, remotely piloted aircraft systems (RPAS), also called drones or unmanned aerial vehicles (UAV), were successfully applied to detect wildlife with imaging sensors, such as RGB and thermal-imaging sensors, with superior detection capabilities to those of human observation. Here, we review studies with RPAS which has been increasingly used in wildlife detection and explain how an RPAS-based high-resolution RGB image can be applied to wild animal studies from the perspective of individual detection and population surveys as well as behavioral studies. The applicability of thermal-imaging sensors was also assessed with further information extractable from image analyses. In addition, RPAS-based case studies of acquisition of high-resolution RGB images for the purpose of detecting southern elephant seals (Mirounga leonina) and shape property extraction using thermal-imaging sensor in King George Island, maritime Antarctica is presented as applications in an extreme environment. The case studies suggest that currently available cost-effective small-sized RPAS, which are capable of flexible operation and mounting miniaturized imaging sensors, and are easily maneuverable even from an inflatable boat, can be an effective and supportive technique for both the visual interpretation and quantitative analysis of wild animals in low-accessible extreme or maritime environments.

Integrating satellite and unmanned aircraft system (UAS) imagery to model livestock population dynamics in the Longbao Wetland National Nature Reserve, China

The collection of field-based animal data is laborious, risky and costly in some areas, such as various nature reserves. Although multiple studies have used satellite imagery, aerial imagery, and field data individually for some animal species surveys, several technical issues still need to be addressed before full standardization of remote sensing methods for modeling animal population dynamics over large areas. This study is the first to model the population dynamics of livestock in the Longbao Wetland National Nature Reserve, China by utilizing yak estimations from Worldview-2 satellite imagery (0.5 m) collected in 2010 and yaks counted in a ground-based survey conducted in 2011 in combination with the animal population structure precisely extracted from UAS imagery captured in 2016. As a consequence, 5501, 5357, and 5510 yaks were estimated to appear in the reserve in 2010, 2011 and 2016, respectively. In total, 1092, 1062 and 1092 sheep were estimated to appear in the reserve in 2010, 2011 and 2016, respectively. The uncertainty of the presented method is also discussed. Primary experiments show that both the satellite imagery and UAS imagery are promising for use in yak censuses, but no sheep were observed in the satellite imagery because of the low resolution. Compared to the ground-based survey conducted in 2011, the UAS image estimate and satellite imagery count deviated in yak quantity by 2.69% and 2.86%, respectively. UASs are a reliable and low-budget alternative to animal surveys. No discernable changes in animal behaviors and animal distributions were observed as the UAS passed at a height of 700 m, and the accuracy of UAS imagery counts were not significantly affected by the short-distance animal movement and image mosaicking errors. The experimental results illustrate the advantages of the combination of satellite and UAS imagery in modeling animal population dynamics.

Lightweight Semantic Segmentation Network for Real-Time Weed Mapping Using Unmanned Aerial Vehicles

The timely and efficient generation of weed maps is essential for weed control tasks and precise spraying applications. Based on the general concept of site-specific weed management (SSWM), many researchers have used unmanned aerial vehicle (UAV) remote sensing technology to monitor weed distributions, which can provide decision support information for precision spraying. However, image processing is mainly conducted offline, as the time gap between image collection and spraying significantly limits the applications of SSWM. In this study, we conducted real-time image processing onboard a UAV to reduce the time gap between image collection and herbicide treatment. First, we established a hardware environment for real-time image processing that integrates map visualization, flight control, image collection, and real-time image processing onboard a UAV based on secondary development. Second, we exploited the proposed model design to develop a lightweight network architecture for weed mapping tasks. The proposed network architecture was evaluated and compared with mainstream semantic segmentation models. Results demonstrate that the proposed network outperform contemporary networks in terms of efficiency with competitive accuracy. We also conducted optimization during the inference process. Precision calibration was applied to both the desktop and embedded devices and the precision was reduced from FP32 to FP16. Experimental results demonstrate that this precision calibration further improves inference speed while maintaining reasonable accuracy. Our modified network architecture achieved an accuracy of 80.9% on the testing samples and its inference speed was 4.5 fps on a Jetson TX2 module (Nvidia Corporation, Santa Clara, CA, USA), which demonstrates its potential for practical agricultural monitoring and precise spraying applications.

Applications of Unmanned Aerial Vehicles in Mining from Exploration to Reclamation: A Review

Over the past decade, unmanned aerial vehicles (UAVs) have been used in the mining industry for various applications from mineral exploration to mine reclamation. This study aims to review academic papers on the applications of UAVs in mining by classifying the mining process into three phases: exploration, exploitation, and reclamation. Systematic reviews were performed to summarize the results of 65 articles (June 2010 to May 2020) and outline the research trend for applying UAVs in mining. This study found that UAVs are used at mining sites for geological and structural analysis via remote sensing, aerial geophysical survey, topographic surveying, rock slope analysis, working environment analysis, underground surveying, and monitoring of soil, water, ecological restoration, and ground subsidence. This study contributes to the classification of current UAV applications during the mining process as well as the identification of prevalent UAV types, data acquired by sensors, scales of targeted areas, and styles of flying control for the applications of UAVs in mining.

Using Deep Learning to Count Albatrosses from Space: Assessing Results in Light of Ground Truth Uncertainty

Many wildlife species inhabit inaccessible environments, limiting researchers ability to conduct essential population surveys. Recently, very high resolution (sub-metre) satellite imagery has enabled remote monitoring of certain species directly from space; however, manual analysis of the imagery is time-consuming, expensive and subjective. State-of-the-art deep learning approaches can automate this process; however, often image datasets are small, and uncertainty in ground truth labels can affect supervised training schemes and the interpretation of errors. In this paper, we investigate these challenges by conducting both manual and automated counts of nesting Wandering Albatrosses on four separate islands, captured by the 31 cm resolution WorldView-3 sensor. We collect counts from six observers, and train a convolutional neural network (U-Net) using leave-one-island-out cross-validation and different combinations of ground truth labels. We show that (1) interobserver variation in manual counts is significant and differs between the four islands, (2) the small dataset can limit the networks ability to generalise to unseen imagery and (3) the choice of ground truth labels can have a significant impact on our assessment of network performance. Our final results show the network detects albatrosses as accurately as human observers for two of the islands, while in the other two misclassifications are largely caused by the presence of noise, cloud cover and habitat, which was not present in the training dataset. While the results show promise, we stress the importance of considering these factors for any study where data is limited and observer confidence is variable.

Using RGISTools to Estimate Water Levels in Reservoirs and Lakes

The combination of freely accessible satellite imagery from multiple programs improves the spatio-temporal coverage of remote sensing data, but it exhibits barriers regarding the variety of web services, file formats, and data standards. Ris an open-source software environment with state-of-the-art statistical packages for the analysis of optical imagery. However, it lacks the tools for providing unified access to multi-program archives to customize and process the time series of images. This manuscript introduces RGISTools, a new software that solves these issues, and provides a working example on water mapping, which is a socially and environmentally relevant research field. The case study uses a digital elevation model and a rarely assessed combination of Landsat-8 and Sentinel-2 imagery to determine the water level of a reservoir in Northern Spain. The case study demonstrates how to acquire and process time series of surface reflectance data in an efficient manner. Our method achieves reasonably accurate results, with a root mean squared error of 0.90 m. Future improvements of the package involve the expansion of the workflow to cover the processing of radar images. This should counteract the limitation of the cloud coverage with multi-spectral images.