Abstract: The identification of cow estrus is an important part of dairy farm production activities and directly related to the economic benefits of dairy farmers. However, there were many problems in present methods for identifying cow estrus, including low efficiency, poor timeliness and low accuracy. To deal with these problems, this paper designed and implemented a real-time monitoring system for cow estrus on the basis of Storm —— a real-time streaming framework in the field of big data. Firstly, cow symptom parameters were obtained from AfiTag Ⅱ made by Afimilk, and then transmitted to the server through the wireless local area network. Secondly, the symptom parameters were processed by Storm. Finally, the data after processing by Storm were visualized on JavaWeb. Taking 2 h as a single time slice, 6 h as a significant sliding window during estrus period, and cow estrus SVM prediction model was established with the vectors including the number of steps s1, s2, s3, cumulative rest time t1, cumulative lying time b and cumulative standing time t2 of three consecutive time units in significant window of estrus. The testing results showed that the average delay of the system was within 2 s, the average accuracy rate was above 98.9%. The prediction accuracy of cow estrus was 85.71%, and cow estrus cycle was shortened to 6 h. The system provided an effective tool for predicting cow estrus, and also had certain guidence for monitoring other large animals.

Key words: cow, estrus information, real-time monitoring, significant window of estrus, Storm, SVM

摘要： 奶牛发情识别是奶牛场生产活动中的重要组成部分，直接关系到奶农的经济效益，而现有的奶牛发情识别方法存在效率低、时效性差、准确率低等问题。针对这些问题，基于大数据实时流式框架Storm设计并实现了奶牛发情实时监测系统。利用阿菲金二代计步器获取奶牛体征参数，通过无线局域网传输到服务器，采用基于Storm的实时流式框架进行处理，JavaWeb对处理后的体征参数可视化展示；以2 h作为单个时间片，6 h作为一个情期显著滑动窗口，选取窗口内连续3个时间片单元的步数s1、s2、s3、累积静卧时间t1、累积起卧次数b和累积站立时间t2为特征向量，建立了基于Storm的奶牛发情SVM预测模型。测试结果表明，设计的系统平均延迟在2 s内，平均准确率在98.9%以上，奶牛发情预测准确率为85.71%，奶牛发情预测周期缩短为6 h。该系统为奶牛发情预测提供了有效工具，对其他大型动物的监测也具有一定的指导意义。

关键词: 奶牛, 发情信息, 实时监测, 情期显著窗口, Storm, SVM