Beyond the Visual Spectrum: From RGB-Based Learning to Hyperspectral Intelligence for Plant Disease...

Area of Science:

• Agricultural Science
• Computer Science
• Remote Sensing

Background:

• Plant diseases cause significant global crop losses, threatening food security.
• Conventional disease detection methods are labor-intensive and often detect symptoms late.
• Hyperspectral imaging (HSI) and deep learning (DL) offer advanced solutions for early and accurate plant disease diagnosis.

Purpose of the Study:

• To provide a comprehensive review of HSI and DL for plant disease detection.
• To analyze the principles, architectures, and challenges of integrating HSI and DL.
• To propose a roadmap for advancing HSI-DL systems from laboratory to field applications.

Main Methods:

• Review of scientific literature from 2008 to 2026 on HSI and DL in plant pathology.
• Analysis of biological and physical principles enabling HSI for plant-pathogen interaction detection.
• Taxonomic classification of DL architectures, including CNNs, hybrid models, and vision transformers.

Main Results:

• HSI captures spectral signatures for pre-symptomatic disease detection.
• DL models, particularly advanced architectures, show high accuracy in controlled environments (95-99%).
• A significant performance gap exists between lab (95-99%) and field (70-85%) data due to environmental variability and domain shift.

Conclusions:

• Future advancements require addressing the lab-to-field performance gap.
• Focus on robustness, scalability, affordability, and interpretability is essential for practical HSI-DL deployment.
• Integrated HSI-DL systems hold promise for enhancing global food security through improved crop monitoring.