Hybrid CNN–Attention Framework for Intelligent Crop Disease Detection and Monitoring

Plant diseases pose a significant threat to sustain-able agriculture and farmer livelihoods by reducing crop yields and increasing input costs. Manual diagnosis is laborious, time-consuming, and often impractical for large farms. In this work, we propose a Hybrid CNN–Attention framework for robust, interpretable, and efficient detection of crop diseases from leaf imagery. The architecture blends convolutional feature extraction with a spatial attention module that highlights disease-specific regions while suppressing background noise. We evaluate the approach on the PlantVillage dataset and additional field-like subsets to assess generalization. The hybrid model achieves strong improvements over baseline CNN sand pretrained backbones in accuracy, precision, recall, and calibration metrics. We include ablation studies, per-class analyses, and deployment considerations for mobile, drone, and edge devices. Attention-based visualizations increase model interpretability and can support human-in-the-Ioop workflows. We also discuss limitations, dataset bias, and directions for multimodal fusion and federated learning for privacy-preserving on-device training.