A method and system for identifying crop lesions based on multispectral images

By employing multispectral imaging technology and evidence fusion methods, the instability of crop lesion identification under changes in light intensity and leaf vein interference was resolved, enabling interpretable output and self-correction capabilities for lesion coverage.

CN122089658APending Publication Date: 2026-05-26SHENYANG AGRI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHENYANG AGRI UNIV
Filing Date
2026-01-27
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing technologies struggle to consistently output interpretable coverage information for crop lesions under conditions of light variation, shadow, and leaf vein interference, leading to delayed prevention and control and inefficient resource allocation.

Method used

Using multispectral imaging technology, a reflectance cube is generated through radiometric calibration and reflectance normalization. Combined with endmember library-constrained spectral unmixing, optimal transmission alignment correction, and evidence fusion, a lesion mask is generated and the coverage interval and interval width are output. The lesion is then identified using deep neural networks and Dempster-Shafer evidence theory.

Benefits of technology

It achieves stable output of lesion masking under complex field conditions, suppresses leaf vein texture and shadow artifacts, provides interpretable boundaries of lesion coverage, and has self-correction capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122089658A_ABST
    Figure CN122089658A_ABST
Patent Text Reader

Abstract

This invention relates to the field of agricultural informatization and intelligent detection technology, specifically to a method and system for identifying crop lesions based on multispectral images. The method includes: acquiring multispectral images of leaves and generating reflectance cubes and leaf masks through radiometric calibration and reflectance normalization; performing constrained spectral unmixing based on an endmember library within the leaf mask region to obtain a lesion abundance map and an unmixing residual map, and generating a corrected reflectance cube by aligning the reflectance cube with optimal transmission; generating a lesion segmentation probability map and leaf vein information based on the corrected reflectance cube, and performing evidence fusion with the lesion abundance map to obtain a lesion confidence map and fusion conflict degree, outputting the lesion mask and lesion coverage interval and interval width; when the fusion conflict degree or interval width exceeds a threshold, acquiring counter-evidence multispectral image data for repeated processing and updating the endmember library, and rolling back when the rollback criterion is met. This invention achieves stable identification and interpretable quantification under complex acquisition conditions.
Need to check novelty before this filing date? Find Prior Art