Plant stress detection using multimodal imaging and machine learning: from leaf spectra to smartphone applications

Area of Science:

• Agricultural Science
• Plant Physiology
• Remote Sensing

Background:

• Spectrophotometry and fluorescence analysis are established methods for detecting plant stress across various spectral regions (UV, RGB, NIR, IR).
• Hyperspectral and multispectral imaging provide detailed physiological insights, enabling earlier stress diagnosis than traditional methods.

Purpose of the Study:

• To review conventional and contemporary sensing approaches for plant stress detection.
• To explore the application of machine learning with multimodal images and smartphone-based diagnostics.
• To identify limitations and future research directions for scalable plant stress models.

Main Methods:

• Utilizing hyperspectral and multispectral imaging for detailed plant physiological assessment.
• Employing thermal and fluorescence imaging to analyze leaf temperature and photosynthetic efficiency.
• Leveraging machine learning for automated feature extraction, classification, and prediction.
• Investigating advancements in digital, thermal, and optical imaging technologies, including smartphone cameras.

Main Results:

• Hyperspectral imaging offers high sensitivity and accuracy but incurs high costs and data demands.
• Multispectral and RGB cameras are more portable and affordable but have lower spectral resolution.
• Emerging low-cost, field-deployable platforms utilizing new imaging technologies enable real-time plant stress monitoring.
• Machine learning integration automates analysis, reducing reliance on expensive equipment and expertise.

Conclusions:

• Advancements in sensing technologies and machine learning facilitate cost-effective, real-time plant stress monitoring.
• Challenges remain in sensor calibration, model generalization across diverse species and environments, and the need for large annotated datasets.
• Smartphone-based solutions show significant potential for inexpensive agricultural diagnostics, though further research is needed for scalable and generalizable models.