Detection and discrimination of disease and insect stress of tea plants using hyperspectral imaging combined with wavelet analysis

Compared with the traditional visual detection method, hyperspectral imaging enables efficient and non-destructive plant monitoring. Besides, it has great potential in plant phenotyping in response to disease and insect infections. However, most previous studies on hyperspectral imaging have focused on detecting a single disease, which can rarely discriminate between multiple co-occurring diseases and insects. In this study, three tea plant stresses with similar symptoms, including the tea green leafhopper (Empoasca (Matsumurasca) onukii Matsuda), anthracnose (Gloeosporium theae-sinesis Miyake), and sunburn (disease-like stress), were evaluated. A multi-step approach was proposed based on hyperspectral imaging and continuous wavelet analysis (CWA) to discriminate the plant stresses. The process entailed: (1) Feature extraction for detection and discrimination of tea plant stresses based on CWA; (2) Detecting abnormal areas on tea leaves via the k-means clustering and support vector machine algorithms; (3) Construction of a model for identification and discrimination of the three tea plant stresses via the random forest algorithm. The results showed that CWA could effectively identify spectral features for distinguishing the three stresses. The overall accuracy (OA) of the proposed approach reached 90.26%-90.69%, with anthracnose having the highest OA (94.12%-94.28%), followed by tea green leafhopper (93.99%-94.20%), while sunburn damage was the least (82.50%-83.91%). Therefore, hyperspectral imaging is effective for plant phenotyping after diseases and insect infections.