基于改进SSD的夜间猪群舒适度监测  

Pig comfort monitoring based on improved SSD at night

作　　者：刘伟[1] 王芳 张苏楠 郭融 LIU Wei;WANG Fang;ZHANG Sunan;GUO Rong(College of Electrical and Power Engineering,Taiyuan Univeristy of Technology,Taiyuan,Shanxi Province 030024,China)

机构地区：[1]太原理工大学电气与动力工程学院,山西太原030024

出　　处：《中国饲料》2020年第14期28-31,共4页China Feed

基　　金：863国家高技术研究发展计划项目(2013AA102306)。

摘　　要：针对圈养猪夜间难以监测的问题,提出一种基于改进SSD的夜间猪群舒适度监测方法。首先,在SSD(single shot multibox detector)算法中引入焦点损失函数,解决夜间目标与背景差异小而难以识别的问题;其次,使用soft-NMS算法检测遮挡的猪只目标;再次,通过二次迁移学习方法得到夜间圈养猪目标检测模型;最后,分析猪只位置信息监测夜间猪群舒适度。结果表明,所提方法能精确检测夜间多只猪目标,并对其舒适度进行监测。Aiming at the problem that it is difficult to monitor pigs at night,a method for monitoring the comfort of pigs at night based on improved SSD is proposed.First,the focus loss function is introduced in the SSD(single shot multibox detector)algorithm to solve the problem that the difference between the pig target and the background at night is difficult to identify.Secondly,the soft-NMS algorithm is used to detect occlusion pig targets.Thirdly,the night pig target detection model is obtained through the secondary transfer learning method.Finally,analyze the location information of pigs to monitor their comfort at night.The results show that the proposed method can accurately detect multiple pig targets at night and monitor their comfort.

关 键 词：多目标检测 焦点损失 软化非极大抑制 二次迁移学习 夜间监控 舒适度 

分 类 号：S828[农业科学—畜牧学]