A Deep Learning Approach to Tomato Disease Classification Using a CNN-LSTM Hybrid Network
Authors:-Youssef Laatiri, Mohamed Ali Mahjoub
Abstract-Our work proposes a classification architecture based on deep learning techniques, particularly convolutional and recurrent neural networks, for the classification of tomato diseases from digital images. More specifically, the objective is to classify leaves infected by a disease using supervised learning on a pre-labeled image dataset from PlantVillage. One of the main challenges of using deep learning, however, is the need for a very large amount of annotated data, which is not always available. Therefore, the objective of our study is to develop a specific hybrid architecture, CNN-LSTM (Convolutional Neural Networks – Long Short-Term Memory), capable of leveraging small (frugal) and relatively imbalanced datasets. To assess the relevance of this approach, we propose to compare it with deep learning algorithms frequently described in the literature. The proposed model achieved better classification performance in terms of validation Accuracy of 94,16%,
