A Study on the Detection of Cattle in UAV Images Using Deep Learning

Barbedo, Jayme Garcia Arnal; Koenigkan, Luciano Vieira; Santos, Thiago Teixeira; Santos, P. M.

doi:10.20944/preprints201912.0089.v1

Cited by 32 publications

(15 citation statements)

References 20 publications

(30 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The cropping can be based on a whole body of an animal [80,117,118], parts (e.g., face, trunk) of an animal [55,107], or areas around facilities (e.g., enrichment, feeder, and drinker) [86,114,119]. As for some large images, cropping them into small and regular pieces of images can reduce computational resources and improve processing speed [56,64,106,120].…”

Section: Image Croppingmentioning

confidence: 99%

“…Mutual reviews among labelers may also be required to minimize the subjective errors [51]. Additionally, to supply models with completed and accurate features of target objects, animals at edges of images were labeled if over 50% of their bodies were visible [120], or images were removed from development datasets if only small parts of animal bodies were presented [107].…”

Section: Data Labelingmentioning

confidence: 99%

“…The ResNet50 [69,76,120,151] and VGG16 [49,76,101,107,120,151] were the most popular architectures used in animal farming (Table 3). The former may be an optimal model balancing detection accuracy and processing speed thus being widely used.…”

Section: Architectures For Image Classificationmentioning

confidence: 99%

“…Because of excellent performance, the abovementioned CNNs were designed as feature extractors and embedded into architectures in the rest of computer vision tasks. [27] First proposal of modern CNN GitHub [155] (PyTorch) [156] Inception family Inception V1/GoogLeNet [35] Increasing width of networks, low computational cost GitHub [157] (PyTorch) [66,76] Inception V3 [158] Inception module, factorized convolution, aggressive regularization GitHub [159] (TensorFlow) [63,76,120] Inception ResNet V2 [160] Combination of Inception module and Residual connection GitHub [161] (TensorFlow) [69,120] Xception [162] Extreme inception module, depthwise separable convolution GitHub [163] (TensorFlow) [120] MobileNet family MobileNet [149] Depthwise separable convolution, lightweight GitHub [159] (TensorFlow) [120] MobileNet V2 [164] Inverted residual structure, bottleneck block GitHub [165] (PyTorch) [120] NASNet family NASNet Mobile [150] Convolutional cell, learning transformable architecture GitHub [166] (TensorFlow) [120] NASNet Large [150] GitHub [159] (TensorFlow)…”

Section: Architectures For Image Classificationmentioning

confidence: 99%

“…The situation is opposite for small batch sizes. Common batch sizes are in a power of 2, ranging from 2 to 2 in animal farming [91,120]. They depend on memory of devices (e.g., GPU) used for training and input image sizes.…”

Section: Batch Sizementioning

confidence: 99%

See 4 more Smart Citations

Practices and Applications of Convolutional Neural Network-Based Computer Vision Systems in Animal Farming: A Review

Huang

Chen

et al. 2021

Sensors

View full text Add to dashboard Cite

Convolutional neural network (CNN)-based computer vision systems have been increasingly applied in animal farming to improve animal management, but current knowledge, practices, limitations, and solutions of the applications remain to be expanded and explored. The objective of this study is to systematically review applications of CNN-based computer vision systems on animal farming in terms of the five deep learning computer vision tasks: image classification, object detection, semantic/instance segmentation, pose estimation, and tracking. Cattle, sheep/goats, pigs, and poultry were the major farm animal species of concern. In this research, preparations for system development, including camera settings, inclusion of variations for data recordings, choices of graphics processing units, image preprocessing, and data labeling were summarized. CNN architectures were reviewed based on the computer vision tasks in animal farming. Strategies of algorithm development included distribution of development data, data augmentation, hyperparameter tuning, and selection of evaluation metrics. Judgment of model performance and performance based on architectures were discussed. Besides practices in optimizing CNN-based computer vision systems, system applications were also organized based on year, country, animal species, and purposes. Finally, recommendations on future research were provided to develop and improve CNN-based computer vision systems for improved welfare, environment, engineering, genetics, and management of farm animals.

show abstract

Section: Image Croppingmentioning

confidence: 99%

Section: Data Labelingmentioning

confidence: 99%

Section: Architectures For Image Classificationmentioning

confidence: 99%

Section: Architectures For Image Classificationmentioning

confidence: 99%

Section: Batch Sizementioning

confidence: 99%

See 3 more Smart Citations

Practices and Applications of Convolutional Neural Network-Based Computer Vision Systems in Animal Farming: A Review

Huang

Chen

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

Wild boar recognition using convolutional neural networks

Silva

Pádua²,

Ogusuku³

et al. 2020

Concurrency and Computation

View full text Add to dashboard Cite

Summary Wild boar (Sus scrofa) is a destructive species of swine. They spread diseases, represent a threat to native species, and destroy natural habitats by destabilizing river banks, thus reducing water flow. The monitoring of populations of wild boars is central to the execution and evaluation of methods to control them. To address this issue, in this article, we retrain and apply four convolutional neural networks (CNNs; AlexNet, VGG‐16, Inception‐v3, and ResNet‐50) to classify different species of “bush pigs” in real‐world footage: two native species of the Brazilian fauna, collared peccary (Pecari tajacu) and white‐lipped peccary (Tayassu pecari), and one invasive species, wild boar (S. scrofa). Results show that CNN can be used to classify animals with very similar behavior and appearance and that ResNet‐50 outperforms all compared CNN in terms of accuracy (98.33%) and the lowest probability of false positives (i.e., native species classified as wild boar).

show abstract

Multispecies detection and identification of African mammals in aerial imagery using convolutional neural networks

Delplanque

Foucher

Lejeune

et al. 2021

Remote Sens Ecol Conserv

View full text Add to dashboard Cite

Survey and monitoring of wildlife populations are among the key elements in nature conservation. The use of unmanned aerial vehicles and light aircrafts as aerial image acquisition systems is growing, as they are cheaper alternatives to traditional census methods. However, the manual localization and identification of species within imagery can be time-consuming and complex. Object detection algorithms, based on convolutional neural networks (CNNs), have shown a good capacity for animal detection. Nevertheless, most of the work has focused on binary detection cases (animal vs. background). The main objective of this study is to compare three recent detection algorithms to detect and identify African mammal species based on high-resolution aerial images. We evaluated the performance of three multi-class CNN algorithms: Faster-RCNN, Libra-RCNN and RetinaNet. Six species were targeted: topis (Damaliscus lunatus jimela), buffalos (Syncerus caffer), elephants (Loxodonta africana), kobs (Kobus kob), warthogs (Phacochoerus africanus) and waterbucks (Kobus ellipsiprymnus). The best model was then applied to a case study using an independent dataset. The best model was the Libra-RCNN, with the best mean average precision (0.80 AE 0.02), the lowest degree of interspecies confusion (3.5 AE 1.4%) and the lowest false positive per true positive ratio (1.7 AE 0.2) on the test set. This model was able to detect and correctly identify 73% of all individuals (1115), find 43 individuals of species other than those targeted and detect 84 missed individuals on our independent UAV dataset, with an average processing speed of 12 s/image. This model showed better detection performance than previous studies dealing with similar habitats. It was able to differentiate six animal species in nadir aerial images. Although limitations were observed with warthog identification and individual detection in herds, this model can save time and can perform precise surveys in open savanna.

show abstract

A Study on the Detection of Cattle in UAV Images Using Deep Learning

Cited by 32 publications

References 20 publications

Practices and Applications of Convolutional Neural Network-Based Computer Vision Systems in Animal Farming: A Review

Practices and Applications of Convolutional Neural Network-Based Computer Vision Systems in Animal Farming: A Review

Wild boar recognition using convolutional neural networks

Multispecies detection and identification of African mammals in aerial imagery using convolutional neural networks

Contact Info

Product

Resources

About