Management and Monitoring of Livestock in the Farm Using Deep Learning

Abstract

Livestock management and monitoring system play a crucial role in farm operations. This paper proposes a system for the management and monitoring of livestock on a farm using deep learning techniques. Traditional methods of monitoring livestock involve manual observation, which can be time-consuming and unreliable. Various systems have been developed, however, there are still challenges existing in present livestock classification and counting, including occlusion, animal overlapping, shadow, etc. To improve all these challenges, this paper presents a monitoring system of livestock under different conditions by the end-to-end deep learning model of You Only Look Once version 5 (YOLOv5). The suggested model conducts feature extraction on the original image with the original YOLOv5 backbone network and detects livestock of different sizes for counting on each anchor frame. Additionally, this model identifies and tracks individual animals The Kaggle dataset collected in real-time containing different animals is used as YOLOv5 relies heavily on data augmentation to improve its detection and tracking performance and validate the proposed system. The scaling, resizing, and manipulation of the splitting dataset are done by the Roboflow application. Additionally, this paper seeks to demonstrate the latest research in utilizing Faster Regional convolutional neural networks (R-CNN) and compare its backbones with the original YOLOv5 backbone. The tensor board graphs from Colab show that this proposed system outperformed other R-CNN, achieving an accuracy of 93% on mAP@_0.5%, making it a promising option for intelligent farm monitoring and managing.