Enhanced Coconut Disease Detection with Lightweight Hybrid VIT-CNN

Technologies Used

Python PyTorch TensorFlow Vision Transformer (VIT) CNN MobileViTv3

Overview

This is an ongoing thesis project focused on improving the early detection of diseases in coconut trees. The core of the project is the design, training, and benchmarking of a lightweight hybrid Vision Transformer (VIT) and Convolutional Neural Network (CNN) model. The goal is to create a model that is both highly accurate and efficient enough for deployment on resource-constrained devices for in-field analysis.

Project Methodology

Design a Hybrid Model: Develop and implement a lightweight hybrid neural network architecture (specifically MobileViTv3) that leverages the strengths of both VITs (for understanding global image context) and CNNs (for extracting local features like textures and edges). This hybrid approach aims to achieve higher accuracy than either model could alone, while maintaining a small footprint.
Build a Specialized Dataset: Create and augment a specialized four-class dataset for coconut diseases. This involves gathering a base of images and then applying data augmentation techniques (e.g., rotation, scaling, flipping) to increase the dataset’s size and diversity, which helps prevent overfitting and improves the model’s ability to generalize.
Benchmark Performance: Rigorously test and validate the model’s performance using metrics like accuracy, precision, recall, and F1-score to confirm its effectiveness for deployment on edge devices.
Improve Early Detection: The ultimate aim is to provide a tool that can help farmers identify diseases earlier, enabling quicker intervention and potentially reducing crop loss.