A deep learning model based on RGB and hyperspectral images for efficiently detecting tea green leafhopper damage symptoms

• Novel method for classifying tea green leafhopper symptoms using RGB and hyperspectral imaging. • The database used is entirely new and includes both RGB and hyperspectral data. • The Two-dimensional discrete wavelet transform (2D-DWT) method significantly improves the accuracy of the model. • Innovative method combining deep learning and traditional machine learning techniques. • Non-destructive monitoring has laid the foundation, not only for tea green leafhopper but also for other pests. The tea green leafhopper, also known as " Empoasca onukii Matsuda ," is a common pest of tea plants that can cause significant economic losses when its damage becomes severe. However, traditional methods of recognizing and classifying the damage symptoms of this pest rely on human visual observation, which is time-consuming, labor-intensive, and subjective. In this study, deep learning models were developed that use both RGB and hyperspectral images to rapidly recognize and classify the damage symptoms caused by the tea green leafhopper. The results showed that both RGB and hyperspectral images can effectively capture the different levels of damage to tea buds caused by the leafhopper. For RGB image classification, models were built using ResNet18, VGG16, AlexNet, WT-ResNet18, WT-VGG16, and WT-AlexNet. Among all these models, WT-VGG16 performed the best, achieving an accuracy of 80.0 %. For hyperspectral image classification, models were built using UVE-SVM, CARS-SVM, SPA-SVM, NONE-SVM, UVE-LSTM, CARS-LSTM, SPA-LSTM, and NONE-LSTM. Among all these models, SPA-LSTM performed the best, achieving an accuracy of 95.6 %. This study demonstrates the potential of using RGB and hyperspectral imaging to accurately and non-destructively monitor the occurrence of tea green leafhopper damage. Such monitoring methods could prove to be efficient in managing this pest.