基于动态时间弯曲的马铃薯干腐病发病期时序高光谱诊断方法

本实验针对马铃薯干腐病潜育期到发病期的诊断方法进行研究，利用时序高光谱对病害发生过程中的病症特征进行分析和提取，并基于时序性特征采用动态时间弯曲（dynamic time warping，DTW）聚类算法对时序关键点进行分析，即对发病期初始点进行诊断。本研究在数据预处理中使用图像阈值分割算法提取动态感兴趣区域，利用概率密度比算法剔除病害光谱异常值，在对比病症的光谱与外观后，发现马铃薯干腐病的光谱具有非单调性特征，再基于该非单调性特征使用高斯核函数的主成分权重系数法进行光谱特征提取。最后基于病害特征，利用模糊聚类方法判定时序关键点，其结果正确率仅为66.7%；针对特征时序性再利用DTW聚类算法判定时序关键点，其结果正确率达94.4%。本实验研究表明基于DTW的时序高光谱诊断方法能对马铃薯干腐病发病期进行有效诊断。

Diagnosis of Symptom Appearance of Potato Dry Rot Disease Using Time Series Hyperspectral Imaging Based on Dynamic Time Warping

In the present study, we aimed to develop a method to diagnose the incubation period and symptom appearance of potato dry rot using time series hyperspectral imaging based on dynamic time warping (DTW). The symptom features during the development of dry rot were analyzed and extracted. The time series key point, namely the start point of symptom appearance was analyzed by DTW clustering algorithm based on the time series characteristics. The threshold segmentation algorithm was used to extract the region of interest (ROI) during the data preprocessing, and the probability density algorithm was applied to eliminate the abnormal spectral data. By comparison of the spectra and appearance of potatoes during the development of dry rot, non-monotonic characteristics were observed in the spectra. Further, the spectral characteristics were extracted by kernel principal component analysis (KPCA). Finally, the time series key point was predicted by using fuzzy clustering model (FCM) based on the symptoms features with an accuracy of only 66.7%. By contrast, the prediction accuracy of DTW on the basis of time series features was as high as 94.4%. This study confirmed that the time series hyperspectral imaging based on DTW could effectively diagnose the symptom appearance of potato dry rot.