A review of deep learning algorithms for computer vision systems in livestock

In livestock operations, systematically monitoring animal body weight, biometric measurements, behavior, feed bunk, and other difficult-to-measure phenotypes is manually unfeasible due to labor, costs, stress. Applications of computer vision are growing in importance systems their ability generate real-time, non-invasive, accurate animal-level information. However, the development a system requires sophisticated statistical computational approaches for efficient data management appropriate mining, as it involves massive datasets. This article aims provide an overview how deep learning has been implemented used livestock, such implementation can be effective tool predict accelerate predictive modeling precise decisions. First, we reviewed most recent milestones achieved with respective algorithms Animal Science studies. Then, published research studies which primary analytical strategy image classification, object detection, segmentation, feature extraction. The great number articles last few years demonstrates high interest rapid across species. Deep systems, Mask R-CNN, Faster YOLO (v3 v4), DeepLab v3, U-Net others have Additionally, network architectures ResNet, Inception, Xception, VGG16 several performance these suggests improved applications faster inference. only fully described implementation. Thus, information regarding hyperparameter tuning, pre-trained weights, backbone, hierarchical structure were missing. We summarized peer-reviewed by tasks (image segmentation), algorithms, species, including identification intake, many others. Understanding principles each application crucial develop operations. Such will potentially major impact on industry predicting real-time phenotypes, could future improve farm decisions, breeding programs through high-throughput phenotyping, optimized data-driven interventions.