Object Detection Method and its Application Based on YOLOv5

In light of the current deficiency in image-based target detection methods in poultry farming and the relatively outdated detection technology, this article introduces the YOLOv5 algorithm based on deep learning to address poultry target detection. It proposes a lightweight convolutional neural network (YOLOv5-CP) specifically designed for target detection in the poultry farming domain. To tackle challenges such as occlusion and a high number of network parameters in poultry target detection, the article enhances the backbone network of the YOLOv5 target detection architecture. It incorporates the FasterNeXt lightweight structure, modifies the original YOLOv5 loss function, applies the ECIOU loss function, and integrates the CA attention mechanism to improve target position information and extract regions of interest. The enhanced network is applied to intelligent breeding in poultry farming, and its performance is verified using a self-made dataset of cage-raised poultry targets. The improved network not only reduces the number of network parameters but also increases detection accuracy by 0.9%.