Optimized Potato Leaf Disease Detection with an Enhanced Convolutional Neural Network

Abstract

Potatoes are one of the most popular and widely consumed vegetable, yet their cultivation faces challenges due to diseases like Early and Late blight, causing significant yield losses. Timely and precise diagnosis of such diseases is critical for adopting effective mitigation approaches and maintaining food security. This research introduces a lightweight deep convolutional neural network for real time potato leaf disease classification. Trained on high-resolution images of healthy and diseased leaves, our optimized model balances accuracy and computational efficiency, making it suitable for resource-constrained devices. The key innovation of our approach lies in the design of an optimized convolutional neural network model, which significantly minimizes the number of trainable parameters by reducing model’s depth, and computational complexity while preserving high classification accuracy. With only 204,227 trainable parameters, it achieves 98.6% test accuracy, along with 98.9% training and 98.6% validation accuracy after 50 epochs . The model demonstrates high precision of 0.99 for early blight, 0.98 for late blight, and 1.00 for healthy potato leaves. A comparative analysis with established architectures like VGG-16, Alex Net, and ResNet-50 underscores the superiority of our approach in terms of performance and computational efficiency. The research provides a significant step forward in leveraging AI for precision agriculture, reducing the impact of potato diseases, and promoting global food security through the adoption of scalable and practical disease detection solutions.

Topics & Concepts