A CNN-based image detector for plant leaf diseases classification.

Area of Science:

• Precision Agriculture
• Computer Vision
• Embedded Systems

Background:

• Accurate plant disease identification is crucial for effective crop management and yield optimization.
• Existing methods often require complex infrastructure or specialized expertise, limiting accessibility in field conditions.

Purpose of the Study:

• To develop and evaluate a resource-constrained convolutional neural network (CNN) for real-time plant disease classification.
• To implement this CNN on a low-cost, low-power embedded platform (OpenMV Cam H7 Plus) for practical field application.

Main Methods:

• A convolutional neural network (CNN) was trained on the ESCA and PlantVillage-augmented datasets for plant disease detection.
• The trained CNN was deployed on an OpenMV Cam H7 Plus, a Python-programmable machine vision camera with an LCD display.
• Real-time image acquisition and classification were performed directly on the embedded device.

Main Results:

• The system achieved high classification accuracies of approximately 98.10% (ESCA dataset) and 95.24% (PlantVillage-augmented dataset).
• The implementation demonstrated efficient performance with low memory usage (around 719-736 KB) and fast inference times (123-126 ms).
• The embedded system proved effective for real-time plant leaf disease classification on the target platform.

Conclusions:

• A portable, embedded system for plant disease classification is feasible using resource-constrained CNNs on low-power platforms.
• The developed solution offers a cost-effective and accessible tool for precision agriculture, aiding farmers in timely disease management.