AI-driven agricultural productivity: detecting potato leaf diseases with multi-scale feature extraction
TL;DR
A unique, computationally efficient triple-feature block network capable of highly accurate plant disease classification across diverse species and complex imaging environments is proposed.
Abstract
Diseases that affect plant leaves significantly reduce crop yields and quality, so it’s essential to detect them promptly and accurately for effective farm management. Deep learning techniques are a strong approach to this issue. We propose a unique, computationally efficient triple-feature block network to address this problem. The approach combines a multiscale feature extraction module that uses parallel dilated depthwise convolutions and SoftMax-based pixel-aware weighting with an image enhancement pipeline that uses LAB color space conversion and Contrast Limited Adaptive Histogram Equalization (CLAHE) to reduce lighting variations. The model was evaluated on a primary dataset of 3,472 images encompassing seven distinct potato leaf conditions captured under unconstrained field settings, and subsequently cross-validated on a secondary dataset of 2,048 medicinal plant images across ten classes to test domain flexibility. The proposed architecture achieved a classification accuracy of 84.30% on the potato dataset and 92.08% on the medicinal plant dataset, outperforming several state-of-the-art deep learning models. Additionally, Grad-CAM was applied to visualize the model’s decision-making process, providing explainable AI insights. Ultimately, this work contributes a robust, lightweight diagnostic framework capable of highly accurate plant disease classification across diverse species and complex imaging environments.
