基于改进YOLOv4的羊只检测方法研究

Sheep detection method in farm environment based on improved YOLOv4

羊只自动检测是大规模智能化羊养殖的基础。针对养殖场环境中存在围栏遮挡以及目标相互遮挡导致检测方法效率低、精度差和易漏检等问题,本文提出了一种基于改进YOLOv4的养殖场环境下羊只检测方法。采用轻量级网络ShuffleNet V2作为主干特征提取网络,使用深度可分离卷积替换普通卷积,在提升检测精度的同时将网络轻量化;引入注意力机制增强特征提取能力;在预测阶段使用DIoU-NMS提高检测精度。改进方法的检测精度达93.57%,检测速度达60 frame/s,参数量降低至41.13MB,能够有效提升养殖场环境下羊只检测的精度与速度。

Automatic detection of sheep is the basis for large-scale intelligent sheep breeding.Aiming at the problems of low efficiency,poor accuracy and easy leakage of detection methods due to fence occlusion in the farm environment and mutual occlusion of targets,this paper proposed a sheep detection method based on improved YOLOv4 in the farm environment.The lightweight network ShuffleNet V2 was used as the backbone feature extraction network,and the ordinary convolution was replaced by depthwise separable convolution,which improved the detection accuracy and lightened the network.Introduction of attention mechanisms to enhance feature extraction;DIoU-NMS was used during the prediction phase to improve detection accuracy.The detection accuracy of the improved method is 93.57%,the detection speed is 60 frame/s,and the parameter amount is reduced to 41.13 MB,which can effectively improve the accuracy and speed of sheep detection in the farm environment.