Advanced multimodal thermal imaging for high-precision fruit disease segmentation and classification

The urgent necessity to bolster agricultural productivity while ensuring quality control has amplified the demand for advanced diagnostic methods for fruit disease detection. Thermal imaging, a promising non-destructive technique, remains underutilized due to complexities and inefficiencies in existing processing models, particularly in handling multiple disease types and maintaining performance at scale. Current methods falter with increased disease variability, presenting a challenge in real-time applications due to their computational intensity and reduced accuracy. Addressing these limitations, this study introduces a robust multimodal analysis framework for fruit disease segmentation and classification based on thermal scans. The proposed model begins with the collection of thermal images of fruits, employing entropy-based Saliency Maps for precise image segmentation. To effectively represent the distinctions of these segmented images, the model harnesses a comprehensive suite of transformations—Frequency, Z Transform, S Transform, and Gabor Transforms—tailoring multi domain features to distinguish between disease states. A pivotal advancement is the integration of Coot Optimization (CO), which streamlines the feature selection process, significantly diminishing redundancy and isolating the most discriminative features for disease identification. Classification is adeptly managed by a novel Graph-based Generative Adversarial Network (Graph GAN) that innovatively combines Graph Neural Networks with the generative capabilities of GANs, offering a powerful blend for categorizing fruit diseases. Upon rigorous testing with mango and apple thermal images, the model demonstrated a remarkable increase in performance metrics, outstripping contemporary methods by achieving a 9.4% enhancement in accuracy, a 4.5% rise in precision, a 3.9% improvement in recall, and a substantial 8.3% reduction in processing delays. The implications of this work are profound, signaling a paradigm shift in agricultural disease management. By significantly elevating the speed and precision of disease detection through thermal imaging, this model paves the way for large-scale, real-time monitoring, potentially revolutionizing fruit disease diagnosis and helping to secure global food supplies in an era of increasing environmental challenges.