학술논문
基于YOLOv5s的母猪基础行为识别 / Sow behavior recognition based on YOLOv5s
Document Type
Academic Journal
Author
Source
安徽科技学院学报 / Journal of Anhui Science and Technology University. 38(1):97-103
Subject
Language
Chinese
ISSN
1673-8772
Abstract
目的:探究机器视觉技术在母猪行为识别中的应用,以及提高遮挡情况下的识别精度.方法:本研究基于YOLOv5s算法,针对母猪的站、坐、躺、爬、趴等 5 种行为,建立母猪行为识别模型.通过使用图像处理技术优化训练数据集,识别模型添加CBAM注意力模块,提高对被遮挡母猪行为的检测精度,最终实现复杂环境下母猪的行为识别,为判断母猪当前状态提供参考.结果:经过优化与反复训练,模型最终检测的精度值较高,达到 97.58％,召回率为 89.69％,单张图片识别时间约为 0.047 s,精确度比未优化前提升了 1.23％.结论:应用YOLOv5s可实现母猪的行为识别,且准确率较高,识别时间较短,识别结果与人工识别结果基本一致,符合猪场实际的养殖要求.
Objective:To explore the application of machine vision technology in sow behavior recognition and improve the recognition accuracy in occlusion situations.Methods:Based on the YOLOv5s algorithm,a sow behavior recognition model was established for five behaviors,which were standing,sitting,lying,crawling and lying on the stomach of sows.By using image processing technology to optimize the training dataset,the recognition model added CBAM attention module to improve the detection accuracy of the behavior of the shielded sow,and finally realized the behavior recognition of the sow in complex environments,which provideed a reference for judging the current state of the sow.Results:After optimization and repeated training,the accuracy of the final detection of the model was high,reaching 97.58％,the recall rate was 89.69％,and the recognition time of a single image was about 0.047 s,which was 1.23％ higher than before optimization.Conclusion:The application of YOLOv5s could realize the behavior recognition of sows,and the accuracy rate was high,the recognition time was short,and the identification results were basically consistent with the manual identification results,which met the actual breeding requirements of pig farms.
Objective:To explore the application of machine vision technology in sow behavior recognition and improve the recognition accuracy in occlusion situations.Methods:Based on the YOLOv5s algorithm,a sow behavior recognition model was established for five behaviors,which were standing,sitting,lying,crawling and lying on the stomach of sows.By using image processing technology to optimize the training dataset,the recognition model added CBAM attention module to improve the detection accuracy of the behavior of the shielded sow,and finally realized the behavior recognition of the sow in complex environments,which provideed a reference for judging the current state of the sow.Results:After optimization and repeated training,the accuracy of the final detection of the model was high,reaching 97.58％,the recall rate was 89.69％,and the recognition time of a single image was about 0.047 s,which was 1.23％ higher than before optimization.Conclusion:The application of YOLOv5s could realize the behavior recognition of sows,and the accuracy rate was high,the recognition time was short,and the identification results were basically consistent with the manual identification results,which met the actual breeding requirements of pig farms.