Fusion‐Based CNN Ensemble with Grad‐CAM for Trustworthy and Transparent Plant Disease Detection

Early and correct detection of plant diseases is absolutely essential for food security, sustainable agriculture, and crop yield optimization. Although deep learning models show great promise in automating disease diagnosis by image-based categorization, their use in actual agricultural environments poses questions about model dependability, openness, and interpretability. This work offers a strong and interpretable deep learning framework for plant disease classification combining improved trustworthiness in decision-making with excellent diagnostic accuracy. To produce physiologically relevant and visually interpretable predictions, the proposed system combines Explainable Artificial Intelligence (XAI) approaches like Gradient-weighted Class Activation Mapping (Grad-CAM) with attention-enhanced Convolutional Neural Networks (CNNs). Training and assessment use a well selected high-resolution dataset comprising several crop species and disease types including grape leaf diseases including downy mildew, powdery mildew, anthracnose, and leaf blight. Comprising thorough preprocessing, focused data augmentation, and transfer learning using cutting-edge CNN architectures including ResNet and EfficientNet, the pipeline Classification robustness is enhanced much further using a fusion-based ensemble learning method. Grad-CAM graphics enable human in the loop validation by highlighting disease-specific lesion sites and hence provide deeper agronomic insights. Experimental results reveal that across different plant species and environmental situations the framework often achieves over 95% classification accuracy. This research provides a scalable, explainable, privacy-conscious AI-powered diagnostic solution that supports openness and dependability in precision agriculture, therefore arming farmers, agronomists, and agricultural extension workers with actionable, reliable insights.