Fighting Grape Black Rot with Deep Learning: A CNN-LSTM Hybrid Model for Disease Severity Classification

Abstract

Grape black rot is a devastating disease that affects grape crops globally. Detecting and preventing the disease as early as possible is crucial for minimizing crop loss and increasing yield and quality. In this study, we propose a CNN-LSTM hybrid model for multi-classification of the severity of grape black rot based on six distinct disease degrees. The model obtained an accuracy of 93.06% after being trained on a dataset of 10,000 grape leaf images collected from an Indian vineyard, outperforming all other deep learning (DL) and traditional machine learning models. The capacity of the proposed model to capture both spatial and temporal characteristics of grape leaf images, as well as the application of data augmentation techniques and early halting, contributed to its superior performance. The model can be utilized as an efficient instrument for the early detection and prevention of grape black rot disease, thereby contributing to the enhancement of the yield and quality of grape crops. However, the model’s performance varied depending on the degree of the disease, with reduced classification accuracy for grape leaves with severe black rot degree. To enhance the model’s ability to precisely classify leaves with severe black rot, additional research is required. Overall, the CNN-LSTM hybrid model proposed for multi-classification of grape black rot severity is a promising approach for the early detection and prevention of grape black rot disease, with potential applications to other plant disease detection tasks.