Based on the Improved Y olov5s Pig Face Detection Research of the Algorithm

In view of the large differences in facial features and low recognition accuracy of pigs under different conditions, this paper proposes a pig face target detection algorithm improved by YOLOv5s, which integrates coordinate CBAM dual-channel attention mechanism in the Backbone network of the model to suppress unnecessary features and pay more attention to key areas of the pig face. The Conv2Former structure was introduced to fully perceive the features and background of pig face, and adapt to the recognition and detection of pig face in different environments. The regression loss function of bounding box is improved, GIoU is replaced by EIoU loss function, the prediction of bounding box is optimized, and the adaptability of the algorithm to pig faces with different scales and shapes is enhanced. Experimental results show that the average accuracy rate of the improved YOLOv5s algorithm reaches 88.6%, which is 6.6% higher than that of the original model, the mAP value is increased from 82.3% to 89.8%, and the detection speed is 64 frames/SEC, which basically meets the requirement of pig face recognition.