Hyperspectral imaging combined with convolutional neural network for outdoor detection of potato diseases

Abstract

Potato is a kind of cultivated crops with a long history. It is the third most important food crop in the world, next to wheat and corn, and plays an important role in agricultural production. Potato late blight (PLB) is one of the main Potato diseases, which can cause 20-40% of the Potato yield loss and bring great losses to the Potato industry. Therefore, it is very important to identify the disease quickly and effectively and take correct control measures. In this study, a new outdoor classification method for potato late blight was proposed, and the outdoor data collected were divided into 5 categories. Due to the small number of experimental samples, a PLB early outdoor detection method combining convolutional neural network (CNN) and spectral data preprocessing method was proposed in this study. They are respectively CNN and multiple scattering correction (CNN-MSC), CNN and wavelet transform (CNN-WT), CNN and first-order difference (CNN-D1), and CNN and second-order difference (CNN-D2). The experimental results show that the accuracy of CNN model test set reaches 97.18%, and the accuracy of CNN-WT and CNN-D2 model test set reaches 100%. The experiment proves the feasibility of using CNN and pretreatment method to detect PLB diseases under the new disease detection method. It provides a new direction for the detection and control of crop pests based on hyperspectral imaging technology.