基于PAM-RF的奶牛活动异常情况监测  

Behavior monitoring of dairy cattle using partitioning around medoids and random forest model(PAM-RF)

作　　者：邓志赟 刘财兴[1] 曹维[1] 尹令[1] 刘汉兴[1] 

机构地区：[1]华南农业大学数学与信息学院,广东广州510642

出　　处：《广东农业科学》2015年第16期122-129,共8页Guangdong Agricultural Sciences

基　　金：广州市科技计划项目(2014Y4300006)

摘　　要：奶牛生理状态信息化监测是实现现代化大规模奶牛养殖的重要工具。针对奶牛的行为模式监测问题,设计了一种利用无线传感器网络技术,基于PAM算法与随机森林算法相结合的奶牛活动异常情况监测模型。该监测模型使用三轴加速度传感器作为奶牛行为数字化采集手段,无监督PAM算法分类样本行为作为训练集,结合基于有监督的随机森林算法作为奶牛活动行为分类的数学模型,在分类奶牛行为基础上构建奶牛活动强度指数时间序列,进而监测奶牛活动异常情况的发生。结果表明,该模型可以高效地分辨高、中、低3类不同强度的奶牛活动行为,模型的平均分类正确率高于91%,其中高强度与低强度分类正确率均高于95%;运用奶牛活动强度指数时间序列能够有效监测奶牛发情异常情况的发生,监测奶牛发情的正确率为91.67%。Intelligent monitoring of cows＇ physiological status is an important tool to achieve the modernization of large-scale dairy farming. According to the problems of monitoring cows＇ behavior, this paper proposed an behavior monitoring model of dairy cattle using wireless sensor networks technology based on the combination of PAM algorithm and random forest algorithm. It used a three-axis accelerometer monitoring as a means to digitally capture the behavior of cows, unsupervised classification algorithm PAM acts as a training set of samples, which combined with supervised random forest-based algorithm acts as a mathematical model of the cow activity classification. And it built a time series of cows＇ behavior index, based on the classification of cows＇ behavior accurately, to monitor abnormal situations of dairy cattle. Experimental results showed that the model could effectively distinguish the high, medium and low intensity of three different cows＇ behaviors. The average classification accttracy rate of the model was above 91%, and both high intensity and low intensity classification accuracy rate reached 95%. And the time series of cows＇ activity intensity index could effectively monitor the occurrence of estrus situations and the accuracy rate of monitoring estrus of dairy cattle reached 91.67%.

关 键 词：无线传感器网络 奶牛 信息化监测 PAM 随机森林 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]