Dairy Cow Behavior Recognition Using Computer Vision Techniques and CNN Networks

Abstract

The application of automated monitoring systems for precision breeding has seen a great increase in recent years. In particular, several studies have addressed the possibility of recognizing cow behavior using computer vision, as well as the opportunity of uniquely identifying and locating individual cows within the barn. In this study, the authors propose a system for recognizing cow behavior within the barn, using a particular type of Convolutional Neural Network (CNN), YOLOv5, and estimation of cattle position via Multi-object recognition. The recordings are obtained from multiple cameras placed inside the barn, a mixed and vast dataset containing several “Cow” objects was obtained and then labeled in two classes “Cow_Standing” and “Cow_Lying.” After the training phase, testing of the network was carried out. The results obtained using this Deep Learning (DL) model, show 94% accuracy, 96% precision and 92% recall in the training phase. In the inference phase, accuracy and recall of 88% and 91% were obtained, respectively.