基于行为学指标数据的猪只行为分类

Pig Behavior Classification Based on Behavior Indexes Data

将视频监控技术应用于规模化养猪场能大大减少人力,提高效率。猪只行为分类是通过图像处理计数从视频中获得猪只信息后的图像信息分析阶段,如何从大量数据中挖掘出猪只的行为是完成实现猪只智能监控的关键。通过图像处理将猪只行为信息提取量化后,生猪的行为识别转化成对行为指标信息的分类问题。首先论述猪只行为识别的关键行为学指标,然后引入决策树分类算法,贝叶斯网络分类算法,基于规则归纳的分类算法等三种分类方法对数据进行实验,并对猪只数据进行分类预测后评估三种模型的表现。结果显示,选取的行为学指标对猪只行为具有较高的区分度,J48决策树分类算法较朴素贝叶斯和基于规则生成的分类算法的准确率均达到96%以上,提取的规则能作为猪只行为分类的判断标准。

The application of visual recognition on large scale pig farm can greatly save manpower and increase efficiency. Pig behavior classification is one of the key problems to be solved for building an effective video surveillance system. Pig behavior recognition can be regarded as the classification of behavior index information via proper classification algorithm. Introduces the key indexes of pig behavior, introduces three kinds of classification methods including C4.5, Bayes Net and RIPPER, and uses the three algorithms to analyze pig behavior data. The resuits analysis indicates that these three algorithm all have an accuracy above 96%, and the J48 which is Weka platform＇s version of C4.5 algorithm, has better performance than the other two methods considering accuracy and time-consuming.