基于模糊最优小波包的植物胁迫因子识别

Plant Stress Recognition Based on Fuzzy-rule Using Optimal Wavelet Packet

为了正确地识别植物常见的胁迫种类,以采集的正常状态和7种胁迫下的植物电信号为样本,结合小波包分解提取特征值能力强的优点,应用模糊准则来优化小波包分解,提取植物电信号中的最优小波包基能量值构成特征集,应用更适合处理模糊的、非线性信号的BP神经网络作为分类器,以实现对不同逆境因子类型的识别。首先利用小波包对采集的植物电信号进行降噪预处理,然后列举了样本经基于模糊准则的小波包处理后各小波包基上的能量样本值,绘制了特征分布图,最后通过对芦荟、碧玉、虎皮兰和蟹爪兰4种植物所处7种胁迫的判断,以统计特征值作为对照,采用所提方法胁迫平均识别率达到95.95%,验证了此方法的准确性和可行性。

In order to correctly identify common stress types of plants, because the wavelet packet has superior ability in feature extraction, the way to optimize the wavelet packet decomposition by use of the fuzzy criterion was proposed. Then BP neural network classifier should be employed to distinguish different stress factors according to physical characteristics of electrical signals performed by plants under diverse stress factors. Because plant electrical signals are very weak, de-noising method based on wavelet packet was used. Afterwards, after wavelet packet was optimized by fuzzy criterion to acquire the feature sets composed of optimal wavelet packet base energy values which can be differentiated easily, electrical signals from plants exposed to a variety of stress types were decomposed by using wavelet packet. In the end, the feature sets were input to a certain BP neural network which is more suited for processing fuzzy and nonlinear signals to finally identify stress types. The results of the study showed that average rate reached to 95.95%. This proposal was proved to be relatively precise and practical after the analysis of four plants including aloe, Jasper, Sansevieria and Schlumbergera, all of which were exposed to seven kinds of detrimental factors.