Automatic detection of broiler’s feeding and aggressive behavior using you only look once algorithm

Abstract

The high market demand for broiler chickens requires that chicken farmers improve their production performance. Production cost and poultry welfare are important competitiveness aspects in the poultry industry. To optimize these aspects, chicken behavior such as feeding and aggression needs to be observed continuously. However, this is not practically done entirely by humans. Implementation of precision live stock farming with deep learning can provide continuous, real-time and automated decisions. In this study, the you only look once version 4 (YOLOv4) architecture is used to detect feeding and aggressive chicken behavior. The data used includes 1,045 feeding bounding boxes and 753 aggressive bounding boxes. The model training is performed using the k-fold cross validation method. The best mean average precision (mAP) values obtained were 99.98% for eating behavior and 99.4% for aggressive behavior.