Efficient deep learning framework for arecanut disease detection using graph neural network and Bat algorithm

Early and accurate detection of plant diseases remains challenging due to real-field variability (e.g., illumination variations, complex backgrounds, occlusion), irregular disease patterns (e.g., subtle or non-local symptoms), limited labeled data for niche crops like arecanut, and generalization issues in uncontrolled environments. Traditional manual inspections are labor-intensive and error-prone, while existing deep learning methods—primarily grid-based CNNs and Vision Transformers—often suffer from limited spatial modeling of non-Euclidean relationships, higher computational costs, and reduced robustness to field conditions. This research introduces a novel deep learning framework for automated detection of diseases in arecanut plants, combining Graph Neural Networks (GNNs) for capturing long-range spatial relationships in leaf images with the Bat Algorithm (BA) for efficient hyperparameter optimization. The framework utilizes a curated balanced subset of 1000 arecanut images from a larger Kaggle dataset (originally 8847 images), captured under natural farm conditions and encompassing nine disease categories. Experimental results demonstrate that the proposed GNN-BA (GB) model achieves 98.45% accuracy, 96.90% precision, 94.21% recall, and 95.05% F1-score—outperforming baselines such as CNN-ViT (93.25% accuracy) and CBAM (94.10% accuracy) by 4–5% on average, while offering lower computational overhead. The model exhibits robustness across diverse environmental and leaf variations, providing a scalable, efficient solution for real-time disease monitoring to enable timely interventions and reduce crop losses in arecanut farming.