摘要
【目的】猪肢蹄病是种猪淘汰的重要原因之一,给养殖场带来巨大的经济损失。猪蹄疾病判断通常依赖人工肉眼观察猪只步态进行排查,存在效率低、人力成本高等问题。本文旨在实现自动化猪步态评分,高效判断猪只肢蹄健康状况。【方法】本文提出一种“端到端”的猪步态评分方法,在单头种猪经过测定通道时采集视频,并制作四分制步态数据集。采用深度学习技术分析视频,设计了一种基于3D卷积网络的时间注意力模块(Time attention module,TAM),有效提取视频帧图像之间的特征信息。将TAM与残差结构结合,构建猪步态评分模型TA3D,对步态视频进行特征提取与步态分类评分。为进一步提升模型性能并实现自动化处理,本文设计了步态关注模块(Gait focus module,GFM),能够自动从实时视频流中提取有效信息并合成高质量步态视频,在提高模型性能的同时降低计算成本。【结果】试验结果表明,GFM可以实时运行,步态视频大小可以减少90%以上,显著降低存储成本,TA3D模型步态评分准确率达到96.43%。与其他经典的视频分析模型的对比测试结果表明,TA3D的准确率和推理速度均达到最佳水平。【结论】本文提出的方案可应用于猪只步态自动评分,为猪肢蹄病的自动检测提供参考。
【Objective】Swine limb and hoof disease is one of the significant reasons for culling breeding swine,resulting in substantial economic losses for livestock farms.The diagnosis of swine limb and hoof disease typically relies on manual observation of pig gaits,which consumes high labor costs and has low efficiency.The aim of this study is to achieve automated pig gait scoring,and efficiently determine the health status of swine limb and hoof.【Method】This study proposed an“end-to-end”pig gait scoring method.Videos of individual breeding swine passing through designated channels were collected and a four-point gait dataset was created.Deep learning techniques were employed for video analysis.A time attention module(TAM)based on a 3D convolutional neural network was designed to effectively extract feature information between video frame images.By combining TAM with residual structures,the pig gait scoring model TA3D was constructed for feature extraction and gait classification scoring in the gait videos.To further improve model performance and achieve automation,the gait focus module(GFM)was designed.GFM could autonomously extract effective information from real-time video streams to synthesize high-quality gait videos,improving model performance while reducing computational costs.【Result】The experimental results demonstrated that GFM could operate in real-time and reduced the size of gait videos by over 90%,significantly reducing storage cost,and the gait scoring accuracy of the TA3D model was 96.43%.Moreover,the comparison test results with other classic video analysis models showed that TA3D achieved optimal levels of accuracy and inference speed.【Conclusion】This paper proposes a solution that can be applied to the automatic scoring of pig gait,providing a reference for the automatic detection of swine limb and hoof disease.
作者
吴振邦
陈泽锴
田绪红
杨杰
尹令
张素敏
WU Zhenbang;CHEN Zekai;TIAN Xuhong;YANG Jie;YIN Ling;ZHANG Sumin(College of Mathematics and Informatics,South China Agricultural University,Guangzhou 510642,China;National Engineering Research Center for Swine Breeding Industry,Guangzhou 510642,China;College of Animal Science,South China Agricultural University,Guangzhou 510642,China;State Key Laboratory of Swine and Poultry Breeding Industry,Guangzhou 510640,China)
出处
《华南农业大学学报》
CAS
CSCD
北大核心
2024年第5期743-753,共11页
Journal of South China Agricultural University
基金
国家自然科学基金(32172780)。
关键词
图像处理
深度学习
猪
步态评分
注意力机制
视频分析
肢蹄病
Image processing
Deep learning
Pig
Gait scoring
Attention mechanism
Video analysis
Limb and hoof disease
