Method for on-line detection of surface defects of cable insulation layer based on machine vision

By using a material-aware perturbation generator and a dual-stream feature decoupling network, the problem of recognition instability of deep learning models in multi-material cable insulation layer detection is solved, achieving stable defect detection in cross-material scenarios and reducing operation and maintenance costs and annotation requirements.

CN122156091APending Publication Date: 2026-06-05广东广缆电缆实业有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
广东广缆电缆实业有限公司
Filing Date
2026-02-09
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing deep learning detection models are unstable in identifying defects in cable insulation layers under different materials, with a high false detection rate. They are also difficult to achieve "one-time training, wide applicability" in multi-material scenarios, requiring additional data collection for new materials, which increases the model's lifecycle cost.

Method used

A spatially adaptive texture perturbation map is generated by a material-aware perturbation generator. The backbone network is decoupled by combining defect-material dual-flow features. Orthogonal loss terms and contrastive mutual information minimization modules are used for decoupling constraints to generate material-invariant defect representations. An online material calibration mechanism is integrated during the deployment phase.

Benefits of technology

It significantly improves the model's ability to perceive the structural invariance of sub-pixel level defects, reduces operational complexity, ensures stable identification of cable insulation defects in multi-material scenarios, and reduces the need for additional labeled data and training frequency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122156091A_ABST
    Figure CN122156091A_ABST
Patent Text Reader

Abstract

The application provides a machine vision-based cable insulation layer surface defect online detection method, comprising the following steps: collecting multi-material defect-free images and labels based on industry standards, and constructing a standardized basic data set; generating spatial adaptive texture disturbance by using a material perception disturbance generator, realizing same defect different material sample enhancement; designing a defect-material dual-flow feature decoupling network, obtaining material-invariant defect representation through orthogonality constraint and mutual information minimization; performing similarity matching between the features and a defect prototype library, realizing defect type and position determination, and online self-adaptive adjustment of noise disturbance intensity, and dynamic updating of the prototype library to maintain the identification performance; and the application can effectively eliminate multi-material interference, realize high-precision automatic identification of insulation layer surface defects, and is suitable for real-time monitoring of a cable production line.
Need to check novelty before this filing date? Find Prior Art