Tomato Plant Disease Detection using TinyML

Description

This project develops a lightweight machine learning model using TinyML techniques to detect tomato plant diseases from leaf images in real-time. By leveraging TensorFlow Lite and Edge Impulse, the model runs directly on edge devices, providing immediate insights for agricultural decision-making. With a dataset of 16,011 images across 10 disease categories, the model undergoes rigorous preprocessing, augmentation, and optimization. Real-world testing validates its effectiveness, offering farmers a convenient tool for on-site disease diagnosis, potentially revolutionizing agricultural practices.

Shortcomings

• Limited Dataset: Despite efforts in augmentation, the dataset may still lack comprehensive representation of environmental factors affecting disease manifestation.
• Dependency on Image Quality: The model's performance could be hindered by variations in image quality, such as lighting conditions and camera resolution, impacting its real-world applicability.
• Dependency of Connectivity: Connectivity limitations in rural areas hinder timely updates and maintenance, compromising the effectiveness of edge-based disease detection models.

Possible Future Extension

• Multi-Crop Expansion: Adapting the model for diverse crops enhances its utility, catering to varied agricultural requirements.
• Mobile App Integration: Creating a user-friendly mobile app enables convenient on-the-go disease diagnosis and intervention for farmers.
• Sensor Data Integration: Incorporating environmental sensor data enhances disease detection accuracy by considering factors affecting plant health.

Dataset Used - Plant Disease

Key Features

• Dataset: Plant Village dataset containing 16,011 tomato leaf images across 10 disease categories.
• Methodology: Data preprocessing, model development, conversion, deployment, evaluation, and validation.
• Results: Achieved accuracy of 89.6% in disease identification.
• Expected Outcomes: High-performing, lightweight ML model, successful deployment on edge devices, real-world effectiveness.
• Real-World Use: Real-time plant health prediction via the Edge Impulse app on mobile devices.

Input Images

Model

Output

Technology Stack

• TensorFlow
• TensorFlow Lite
• Edge Impulse
• Python

How to Use

• Clone the repository git clone https://github.com/its-kumar-yash/Tomato-Plant-Disease-Detection-Model.git
• Download the dataset.
• Install required dependencies.
• Run the jupiter notebook.
• Upload tf_lite_quantized_model.tflite file on Edge Impulse and Build the model.