Textile fabric defect hyperspectral image detection method and system

By using a multi-path neural network with a three-way parallel structure, combined with mirror filling and feature fusion, the problem of insufficient spatial and spectral feature capture in textile fabrics by existing hyperspectral detection methods is solved, achieving high-precision and robust defect detection, which is suitable for online inspection in the textile industry.

CN122335675APending Publication Date: 2026-07-03JIANGSU JICUI FUNCTIONAL MATERIALS RES INST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
JIANGSU JICUI FUNCTIONAL MATERIALS RES INST CO LTD
Filing Date
2026-03-16
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing hyperspectral defect detection methods struggle to simultaneously and efficiently capture spatial context and spectral distinguishing features in textiles, resulting in insufficient accuracy and poor robustness in detecting minute or low-contrast defects, especially under complex texture backgrounds.

Method used

A multi-path neural network with a three-way parallel structure is used to extract global space, local space-spectral collaboration and spectral features respectively, and perform feature fusion, including mirror filling strategy, multi-scale texture feature extraction, local space-spectral correlation features and deep spectral nonlinear features, to generate a comprehensive feature vector for defect discrimination.

Benefits of technology

It significantly improves the accuracy and robustness of detecting minute defects against complex texture backgrounds, making it suitable for industrial online inspection and enabling high-precision defect classification and alarm decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122335675A_ABST
    Figure CN122335675A_ABST
Patent Text Reader

Abstract

The application discloses a textile fabric defect hyperspectral image detection method and system, which comprises the following steps: acquiring a hyperspectral image of a textile fabric and generating a hyperspectral dataset, and pre-processing the hyperspectral image in the dataset; expanding the hyperspectral image based on a mirror filling strategy, and extracting a three-dimensional Patch block set centered on a pixel according to a preset patch size on the expanded image; inputting the Patch block into a multi-branch neural network model, and respectively extracting multi-scale texture features, local spatial-spectral correlation features and spectral deep nonlinear features; performing feature fusion on the outputs of the three branches and defect discrimination, and outputting corresponding defect category probabilities; reconstructing a fabric defect distribution map according to the defect discrimination results of the center pixels of the Patch blocks and performing filtering processing to obtain a final textile fabric defect detection result. The application is aimed at special extraction and fusion of global spatial features, local spatial-spectral collaborative features and pure spectral features, and high-precision and strong-robustness defect detection is realized.
Need to check novelty before this filing date? Find Prior Art