HyperAgriNet: An Explainable Hypergraph Attention Network for Plant Disease Classification in Smart Agriculture

Abstract: Plant diseases are a major cause of reduced agricultural productivity and crop quality, leading to significant economic losses and posing challenges to global food security. Accurate and timely disease identification is essential for effective crop management and precision agriculture. Although deep learning-based approaches have shown excellent performance in plant disease classification, conventional convolutional neural networks (CNNs) mainly focus on local feature extraction and often fail to capture complex relationships among disease symptoms. Moreover, their black-box nature limits interpretability and user trust.
To address these limitations, this paper proposes HyperAgriNet, an Explainable Hypergraph Attention Network for plant disease classification. The proposed framework combines multi-scale feature extraction, lesion-aware attention mechanisms, and hypergraph neural learning to capture both local disease characteristics and higher-order contextual relationships. Multi-scale convolutional layers extract features at different spatial resolutions, while channel and spatial attention modules enhance disease-affected regions. A Hypergraph Neural Network (HGNN) further models complex inter-class relationships, and Grad-CAM is integrated to provide visual explanations by highlighting infected leaf regions.
Experimental results demonstrate that HyperAgriNet achieves an overall classification accuracy of 99.18%, outperforming several state-of-the-art models, including ResNet-50, DenseNet-121, EfficientNet-B3, Vision Transformer, and Swin Transformer. By combining high classification accuracy with explainable predictions, HyperAgriNet offers a reliable and interpretable solution for intelligent plant disease diagnosis in smart agriculture.

Keywords: Plant Disease Classification, Smart Agriculture, Hypergraph Neural Network, Attention Mechanism, Explainable AI, Deep Learning.