基于改进YOLOv10n模型的牛脸识别方法研究OA
Study of a cattle face recognition method based on an improved YOLOv10n model
为了解决常规深度学习模型牛脸识别准确率不高的问题,本研究提出了一种改进的牛脸识别方法——YOLO-SRC 模型,在 YOLOv10n 模型中引入 SPPF_SHViT 模块增强特征提取能力,使用 C2f_RepViT 模块优化特征表示,嵌入 CLA 注意力机制形成 PSA_CLA 模块实现自适应多尺度特征融合.结果表明:YOLO-SRC 模型在华西牛牛脸数据集的平均精度均值(mAP)为 93.6%,精确度为 93.4%,召回率为 83.2%,分别比 YOLOv10n 模型提升了 2.9,2.0,3.5 百分点;与 Faster R-CNN、YOLOv7、YOLOv8n、YOLOv10n 和 YOLOv11n 模型相比,YOLO-SRC 模型的热力图更能关注脸部区域,其置信度也最高.说明 YOLO-SRC 模型具有良好的识别性能与鲁棒性.
In order to solve the limited accuracy of conventional deep learning models in cattle face recognition,YOLO-SRC,an enhanced version of YOLOv10n was proposed.Introducing the SPPF_SHViT module into the YOLOv10n model could enhance feature extraction capability.By using the C2f_RepViT module to optimize feature representation and embedding the CLA attention mechanism to form the PSA_CA module,adaptive multi-scale feature fusion was achieved.The results showed that in the Huaxi cattle face dataset,YOLO-SRC achieved an average precision mean value(mAP)of 93.6%,with an accuracy level of 93.4%,and a recall rate of 83.2%,which were 2.9,2.0,and 3.5 percentage higher than those of YOLOv10n,respectively.Compared with Faster R-CNN,YOLOv7,YOLOv8n,and YOLOv11n,the heatmap results of the YOLO-SRC model showed that it could better focus on facial regions and had the highest confidence,indicating that the YOLO-SRC model had good recognition performance and robustness in cattle face recognition.
何氽;郭鹏;朱波
天津农学院 计算机与信息工程学院,天津 300392天津农学院 计算机与信息工程学院,天津 300392中国农业科学院 北京畜牧兽医研究所,北京 100193
农业科技
牛脸识别深度学习YOLO-SRC性能华西牛
cattle face recognitiondeep learningYOLO-SRCperformanceHuaxi cattle
《黑龙江畜牧兽医》 2026 (6)
63-68,6
国家自然科学基金项目(32272843)甘肃省科技计划资助项目(26CXNM002)中国农业科学院委托项目"牛脸识别系统设计开发"(TNHXKJ2023037)
评论