Research on chicken basic behavior recognition method based on YOLOv5x
The healthy breeding of chickens is of great significance to improve their production and quality.Traditional breeding models lack professional breeding guidance and timely disease prevention and control,and cannot meet the requirements of healthy breeding.Therefore,it is necessary to identify the behavior of chickens to ensure their healthy growth.At present,the behavior identification of chickens in the breeding process mostly adopts manual observation or electronic tags,which has the disadvantages of strong subjectivity and time-consuming and labor-intensive.In view of the problems of similar shapes and actions of chickens in free-range breeding mode,such as easy occlusion between each other,a YOLOv5x-Swin-TransformerV2-SPPF model based on the improvement of YOLOv5x was proposed.On the basis of establishing a chicken behavior recognition dataset by using data augmentation and other technologies,the model achieves automatic recognition of four basic behaviors of chickens,including standing,feeding,lying down,and drinking water.The average precision rates of each behavior are 90.8%for standing,84.3%for feeding,92.8%for lying down,and 91.5%for drinking water.The mean average precision(mAP)of four behavior recognition is 89.9%,which is 9.5%,4.37%,and 3.32%higher than that of YOLOv3,YOLOv5s,and YOLOv5x before improvement,respectively.Through result analysis,the effectiveness of the improved model was verified.It achieves real-time monitoring of chicken behavior activities by using deep learning technology for chicken behavior recognition.It was of great significance to achieve the goal of healthy breeding and sustainable development.
deep learningchicken behavior recognitionYOLOv5xsmall target detection
NETL
NSTL
万方数据
