基于快速SSD的猪群舒适度监测  

Pig comfort monitoring based on fast SSD

作　　者：谢涛 王芳 田建艳 刘伟[1] 郭融 XIE Tao;WANG Fang;TIAN Jianyan;LIU Wei;GUO Rong(College of Electrical and Power Engineering,TaiyuanUniveristy of Technology,Taiyuan 030024,China)

机构地区：[1]太原理工大学电气与动力工程学院

出　　处：《畜牧与兽医》2019年第12期40-45,共6页Animal Husbandry & Veterinary Medicine

基　　金：国家高技术研究发展计划-863计划(2013AA102306)

摘　　要：在自由栏猪舍养殖下,生猪聚散程度可以反映猪群在当前环境下是否处于舒适的生活状态。然而猪舍环境光线变化范围大、生猪为非刚体以及多只生猪之间相互粘连、遮挡等因素会造成多只生猪目标实时识别较为困难。本文以现场生猪监测视频为数据源,提出了一种基于困难样本的快速单网络目标检测器(single shot multibox detector SSD)猪群目标识别算法,并通过分析猪群聚散度,监测当前猪群在猪舍内的生活舒适度。实验结果表明,本文提出的方法简化了SSD网络结构,可以准确识别猪舍中多只生猪目标,识别速度快,满足生猪监测实时识别要求,监测猪群的生活舒适度,有助于提升生猪自动化养殖预警能力。In the free-bar farm,the degree of pigs gathering can tell whether they feel comfort in the current environment.However,the huge range of environmental light changes in pig houses,the non-rigid body of pigs,and the adhesion and occlusion of multiple pigs may cause problems in real-time identification of multiple pig targets.Based on the on-site monitoring video used as the data source,a fast SSD multiple pig targets recognition algorithm based on difficult samples is proposed here.By analyzing their convergence,the comfort of the pigs in the house is monitored.The experimental results showed that the method proposed in this study simplified the SSD network structure which accurately identified multiple pig targets and had a fast recognition speed,meeting the requirement for real-time identification of pig targets to monitor the comfort of pigs.This monitoring technique would be helpful in early warning in automation pig breeding.

关 键 词：多目标识别 舒适度 快速SSD 自主困难样本 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]