Method of Classified Counting of Mixed Breeding Chickens Based on YOLOV5
Erjie Sun, Xiao Zhang, Fengwei Yuan, Zhicheng Wang, Guoning Ma, Junjie Liu
Abstract
Accurately identifying and calculating the number of chickens in a large-capacity mixed-raising environment is a key step in applying machine vision technology to automatically monitor the behavior, status, health, and other precise animal husbandry. This paper proposes a chicken counting method based on YOLOV5, taking the Silky chicken and Xianghuang chicken mixed in a chicken house as recognition targets. Use the camera to record the situation of the mixed chickens before and after entering and leaving the chicken house from 17:00 every day to 7:00 the next day for a week, screen out 321 pictures, mark the Xianghuang chicken (17543 samples) and the Silky chicken (8905 samples), and use the YOLO series of target detection algorithms to identify. The experimental results showed that the yolov5L algorithm had the best effect, and the precision rate of Xianghuang chickens was 89.9% and the recall rate was 87.2%, while the precision rate of black chickens was 85.1% and the recall rate was 84.2%. This method can replace the manual monitoring of the number of chickens returned to the nest, providing the possibility of precise breeding.