Multimodal Image-Based Estimation of Wheat Waterlogging Stress Using Hyperspectral, Thermal and Fluorescence Features

Le résumé fourni par la source

Multimodal imaging has emerged as an effective approach for non-destructive crop stress analysis; however, the relative capability of different image modalities to represent stress-related information and the benefits of their integration remain insufficiently quantified. In this study, wheat waterlogging stress is used as a representative case to investigate hyperspectral, thermal infrared, and chlorophyll fluorescence image features from the perspective of information representation and feature fusion, rather than stress index construction. Single-modality and multimodal feature-based estimation models were developed to evaluate the contribution of each image modality and the performance gains achieved through feature fusion. The results show that hyperspectral features provide dominant representation capability, while thermal infrared and chlorophyll fluorescence features offer complementary information in the feature space. Multimodal feature fusion substantially improves estimation accuracy and robustness. These findings highlight the effectiveness of multimodal image processing strategies for crop stress estimation and provide methodological insights for feature-level integration in agricultural imaging applications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.