Estimating Contents of Multiple Biomarkers in Medicago truncatula Under Salt Stress via Multi-Granularity Spectral Segmentation

The accurate and non-destructive estimation of plant physiological and biochemical indicators (biomarkers) is crucial for crop stress assessment. However, conventional spectral preprocessing methods may not fully exploit the diverse spectral components associated with biomarkers with different spectral response characteristics. To address this issue, we proposed the multi-granularity spectral segmentation (MGSS)-based retrieval framework for the estimation of multiple biomarkers of Medicago truncatula under salt stress. Multi-granularity segmentation (MGS), as one part of the MGSS framework, was used to extract spectral features at different granularities. In this study, we obtained a dataset of leaf spectra and corresponding biomarkers (including relative chlorophyll content (SPAD), soluble sugar (SS), and malondialdehyde (MDA)) for 720 pots of Medicago truncatula. These plants were subjected to different levels of salt stress (including 100, 200, and 250 mmol L⁻¹ NaCl and untreated (CK)). Using PLSR, we compared MGS features with conventional spectral features for biomarker estimation. The results showed that MGS achieved competitive estimation performance on SPAD, SS, and MDA, with the optimal validation R² values of 0.98, 0.69, and 0.36, respectively, compared to the optimal conventional features. More importantly, the optimal granularities were different among the three biomarkers. SPAD, SS, and MDA reached the optimal performance at G2, G5, and G18, respectively, which showed different spectral responses to different frequency levels. This study provides a promising framework for accurately estimating multiple biomarkers in crops under abiotic stress. This method may provide a new interpretable perspective on the differences in spectral retrieval performance among various biomarkers.