This study proposes a yak face recognition model based on YOLOv5s_MobileNetv3_CoordConv.By re-placing the backbone network of YOLOv5s with MobileNetv3 and substituting the standard convolutions in the neck network with coordinate convolutions,the model significantly improves performance while maintaining lightweight characteristics.On a dataset of 14 476 images from 20 yaks,the model achieved an accuracy,recall,and mean average precision(mAP_0.5)of 93.4%,94.9%,and 98.2%,respectively,representing improvements of 3.4%,3.5%,and 1.4%over the YOLOv5s model.Additionally,the model can recognize multiple yaks in a single image,enabling multi-target recognition tasks.
维普
万方数据