A Hybrid Deep Learning Architecture with Advanced Feature Optimization for Apple Leaf Disease Classification

Global apple farming regions are economically and agriculturally impacted by apple leaf diseases. Agricultural disease identification by human observation is laborious, subjective, and ineffective for large-scale application. For accurate apple leaf disease classification, this study proposes an advanced hybrid deep learning network named AHD-Net. In the proposed architecture, hybrid convolutional neural networks (CNNs) extract features at multiple scales, channel-spatial attention refines them, and a novel module optimizes them via mathematical regularization. AHD-Net is applied to the Apple Leaf Disease Dataset from Plant City 2025, and results show that it outperforms leading deep learning models in classification accuracy$(98.7 \%)$and computational efficiency. Ablation tests show that feature optimization reduces feature redundancy by 37.2 % while keeping discriminative disease patterns. Early disease identification is required in precision agriculture, and this study works in this direction. It can reduce agricultural losses and maximize resource use.