Hyperspectral imaging: a novel approach for plant root phenotyping

Area of Science:

• Plant science
• Agricultural engineering
• Spectroscopy

Background:

• Root system architecture is crucial for plant resource acquisition.
• Current RGB imaging methods struggle with poor contrast in soil.
• Hyperspectral imaging is hypothesized to improve root segmentation and provide physico-chemical data.

Purpose of the Study:

• To establish fundamentals for root phenotyping using hyperspectral imaging.
• To develop an image processing pipeline for automated segmentation of soil-grown roots.
• To explore spectral signatures for physico-chemical root properties.

Main Methods:

• Scanning of Triticum durum root systems in soil-filled rhizoboxes (1000-1700 nm, 222 bands, 0.1 mm resolution).
• Development of a data processing pipeline for automatic root segmentation and spectral signature analysis.
• Application of log-linearized and asymptotic least squares correction, fuzzy clustering, and multilevel thresholding for segmentation.

Main Results:

• Spectral and RGB imaging showed similar segmentation accuracy, even with bright soil.
• Optimal spectral segmentation was achieved using advanced image correction and clustering techniques.
• Distinct spectral signatures were observed between root center and border regions, indicating potential for detailed analysis.
• Root decay was quantifiable using spectral differences related to water and carbon absorption.

Conclusions:

• Established fundamentals for hyperspectral imaging in root phenotyping.
• Developed a high-resolution, automated image processing pipeline for soil-grown roots.
• Demonstrated the potential of spectral signatures to encode physico-chemical root properties.