<p>This study addresses the challenge of classifying low-contrast leaf images, particularly for detecting grape leaf diseases. Although convolutional neural networks (CNNs) have demonstrated strong performance in image classification, their effectiveness diminishes when processing low-contrast images. To address this limitation, a novel image enhancement (IE) technique based on the Atangana–Baleanu fractional-order derivative in the Riemann sense (ABR) is proposed. Unlike traditional enhancement methods, the ABR-based convolution kernel offers non-local and non-singular memory properties via the Mittag-Leffler function, enabling effective edge and texture enhancement while preserving smooth regions. This is the first work to formulate the ABR operator as a digital convolution filter for integration into CNN preprocessing. In addition, we introduce a hybrid framework that combines the enhanced images with a GrNet-18 CNN architecture, whose initial kernel parameters are optimized using a genetic algorithm (GA). This dual approach not only improves the visibility of critical features in low-contrast images but also boosts the classification performance. Experimental results show that the proposed method significantly outperforms several state-of-the-art enhancement and classification techniques, achieving an accuracy of 99.03%, recall of 98.82%, specificity of 99.65%, precision of 99.20%, and an F1-score of 99.00%. These results demonstrate the effectiveness and novelty of the proposed approach for the classification of low-contrast leaf disease images.</p>