Soybean high temperature tolerance grading method based on vegetation index prior and self-supervised learning

CN121982552BActive Publication Date: 2026-07-21ANHUI AGRICULTURAL UNIVERSITY

View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Patents(China)
Current Assignee / Owner: ANHUI AGRICULTURAL UNIVERSITY
Filing Date: 2026-04-03
Publication Date: 2026-07-21

AI Technical Summary

Technical Problem

In the early diagnosis of high temperature stress in soybeans, existing technologies rely on single physical vegetation indices, which are easily affected by soil background and canopy structure, making it difficult to accurately capture subtle phenotypic changes. Furthermore, supervised learning methods rely on high-quality labeled data, which is costly and leads to overfitting and insufficient robustness of the models.

Method used

A method based on vegetation index prior and self-supervised learning is adopted. By calculating the mapping relationship between vegetation index and high temperature resistance level, a saliency map is generated. A non-uniform masking strategy is constructed. Combined with a cross-modal cross-attention module and a weighted gating loss module, the model parameters are optimized and robust high temperature resistance texture features are extracted.

Benefits of technology

It achieves high-precision and robust phenotypic identification of soybean under high-temperature stress, improves the efficiency and accuracy of resistance identification of the model, reduces dependence on labeled data, and enhances feature fusion capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN121982552B_ABST

Patent Text Reader

Abstract

The application provides a soybean high-temperature tolerance grading method based on a vegetation index prior and self-supervised learning, wherein an SVM is used to fit the mapping relationship between the vegetation index and the high-temperature tolerance grade, SHAP explainability analysis is introduced to calculate the contribution weight of each vegetation index, and a vegetation index prior significance map reflecting the physiological importance of leaves is generated; the vegetation prior significance map is mapped to an RGB image coordinate system to construct a non-uniform mask strategy, a mask autoencoder is guided to preferentially reconstruct a high physiological significance area, and robust high-temperature tolerance texture features are extracted; according to the features extracted from the RGB image and the multispectral image, a feature vector is constructed, a cross-modal cross-attention module is used to fuse the feature vector, a weighted gate loss module based on a contrast loss and a cross-entropy loss is used to calculate the error, and the front-end network parameters are updated by using back propagation. The application effectively reduces the artificial labeling cost and realizes the identification of the high-temperature tolerance phenotype of a large-scale soybean germplasm.

Need to check novelty before this filing date? Find Prior Art