Hybrid CNN and Transformer-Based Sequential Learning Techniques for Plant Disease Classification

This project focuses on developing an image classification system for identifying plant diseases using deep learning. The system integrates Convolutional Neural Networks (CNNs) with sequential learning models such as BiGRU, BiLSTM, and LSTM. Attention mechanisms and Squeeze-and-Excitation (SE) modules are incorporated to enhance feature learning. Four hybrid models are implemented: EfficientNet with BiGRU and Attention, ResNet50 with BiLSTM and Attention, DenseNet121 with CNN-LSTM, and MobileNetV3 with LSTM and SE Attention. The PlantVillage dataset, which includes 34 distinct plant disease categories, is used for training and evaluation. The system is deployed as a web application built using Flask, with a simple user interface for image upload and classification result display. The combination of advanced CNNs and sequential learning enables improved accuracy and contextual understanding of image features. The project aims to explore the effectiveness of hybrid deep learning techniques in the classification task and provide a working prototype through a functional web interface.

Keywords: CNN, BiGRU, BiLSTM, LSTM, Attention, SE block, EfficientNet, ResNet50, DenseNet121, PlantVillage