基于半监督模糊聚类算法的奶牛行为判别系统

以实时判别奶牛行为,提升精细养殖技术水平为目标,本试验以系统功耗低、检测灵敏度高、运行稳定性强为原则设计无线传感节点,研发了一种基于半监督模糊聚类算法的奶牛行为实时判别系统。为获取最佳通信距离及最优节点固定高度,对无线传感节点分别进行固定高度-通讯距离与丢包率关系测试、固定高度-数据波动关系测试,通信距离分别取10、20、30 m,固定高度分别取10、20、30 cm;并将半监督模糊聚类判别算法、K-means算法及BP神经网络算法在奶牛行为识别方面的准确度、精度及敏感度进行比较。结果显示,集成三轴加速度传感器ADXL345、处理器MSP430-F149、无线收发器CC1101等芯片设计的无线传感节点可精确采集奶牛运动加速度数据,满足长期可靠传输数据等工作要求。固定高度-通讯距离与丢包率、固定高度-数据波动关系测试结果显示,最优传输距离为10 m,最佳节点固定高度为30 cm。半监督模糊聚类算法性能最高,平均准确度达到95.4%,平均精度为53.0%,平均敏感度为60.6%。K-means算法的平均准确度达到90.3%,平均精度仅有39.9%,平均敏感度为45.6%。BP神经网络算法平均准确度达到93.7%,平均精度为45.5%,平均敏感度为47.0%。半监督模糊聚类算法具有准确性高、学习复杂度低、运行速度快的特点,具有良好的寻优能力,效率较高。

A Cow Behavior Classification System Based on Semi-supervised Fuzzy Clustering Algorithm

In order to determine the behavior of cows and improve the management level of fine feed,a Real-time identification system for cows' movement behaviors based on the semi-supervised fuzzy clustering algorithm was developed and the wireless sensor node was designed on the principle of low power consumption,high sensitivity and high operational stability.At the same time,the transmission performance of wireless sensor nodes was tested to determine the optimal transmission distance and the optimal node fixed height,in which the transmission distance were set at 10,20 and 30 m,while the fixed height were 10,20 and 30 cm,respectively.And the accuracy,precision and sensitivity of K-means algorithm,BP neural networks algorithm and semi-supervised fuzzy clustering algorithm were compared.The results showed that the wireless sensor node consist with three-axis accelerometer ADXL345,processor MSP430-F149,and wireless transceiver CC1101 which could collect data automatically of cows' movement behaviors,and meet the long-term reliable data transmission and other work requirements.The test results showed that the optimal transmission distance was 10 m and the optimal fixed node height was 30 cm.The accuracy,precision and sensitivity of semi-supervised fuzzy clustering algorithm were 95.4%,53.0% and 60.6%,respectively,that of K-means algorithm,BP neural network were 90.3%,39.9%,45.6% and 93.7%,45.5%,47.0%,respectively.In conclusion,the accuracy of semi-supervised fuzzy clustering algorithm was high,implementation was simple,learning complexity was low and running speed was fast,and it also had strong optimization ability.