一种新的基于Hu不变矩的猪只姿态识别方法  

A new method of pig posture recognition based on Hu invariant moments

作　　者：王海涛 王芳[1] 田建艳[1] 张聪 

机构地区：[1]太原理工大学信息工程学院,太原030024

出　　处：《黑龙江畜牧兽医》2017年第12期14-17,285,共5页Heilongjiang Animal Science And veterinary Medicine

基　　金："863"国家高技术研究发展计划项目"畜禽养殖数字化关键技术与设备开发"(2013AA102306)

摘　　要：为了实现猪只异常行为的自动监控,试验针对非刚体的猪只姿态进行分类识别,首先将原始图像进行预处理与优化提取所有图像中的7个不变矩,在此基础上对数据进行处理,求出站、坐、趴、躺4种姿态的模板图像与待测图像的不变矩相关系数,最后将所提取的相关系数作为支持向量机的特征向量进行多姿态分类,实现4种姿态的识别。结果表明:该方法的识别率超过90%。说明该方法具有可行性,可用于对猪只进行异常监控。In order to realize the automatic monitoring on pig＇ s abnormal behavior, classified recognition of non - rigid pig posture was carried out in this experiment. Firstly, the original images were preprocessed and the seven invariant moments in each image were optimized and extracted. Based onthe results, the data were processed, and correlation coefficients of the invariant moments of four kinds of postures （ stand, sit, grovel, and lie）were determined between template images and measured images. Finally, extracted correlation coefficients were used to be feature vectors of support vector machine for multi - posture classification, and the recognition of four kinds of postures were realized. The results suggested that recognition rate of this method was more than 90% ,which declared that this method was feasible and could be used for the monitoring on pig＇ s abnormal behavior.

关 键 词：HU不变矩 支持向量机 轮廓提取 姿态识别 特征提取 相关系数 

分 类 号：S828[农业科学—畜牧学] S854.9[农业科学—畜牧兽医]