A zero-shot insulator defect detection method

CN122391112APending Publication Date: 2026-07-14SKILL TRAINING CENT STATE GRID JIBEI ELECTRONICS POWER COMPANY +2
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SKILL TRAINING CENT STATE GRID JIBEI ELECTRONICS POWER COMPANY
Filing Date
2026-04-14
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing technologies lack the ability to accurately identify and locate insulator defects during power line inspections. In particular, under zero-sample or few-sample conditions, traditional methods cannot effectively embed physical priors, eliminate environmental noise, and achieve accurate location.

Method used

A method combining multimodal feature extraction and expert prompt enhancement, spatial-frequency domain dual-stream noise decoupling, frequency domain periodic topology extraction, master-slave game residual correction, and sparsity constraints is adopted. This method is combined with a visual-language model to embed the physical structural features of the insulator, eliminate noise interference, and achieve accurate positioning.

Benefits of technology

It significantly improves the accuracy and positioning precision of insulator defect detection in complex environments, reduces the dependence on data annotation, and enhances the model's anti-interference ability and robustness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122391112A_ABST
    Figure CN122391112A_ABST
Patent Text Reader

Abstract

The application discloses a kind of zero sample insulator defect detection methods, comprising: multi-modal feature extraction and expert prompt enhancement;Space-frequency domain double-flow noise decoupling, through macroscopic and microscopic double-flow noise library and gate network, background interference is adaptively rejected;Frequency domain periodicity topology extraction, based on FFT frequency domain transformation captures insulator umbrella skirt periodicity physical characteristics and is mapped as topological feature vector;Master-slave game residual correction, through adaptive residual coefficient realizes feature smooth fusion, avoids model collapse;Sparsity constraint and weak supervision positioning, utilize L1 sparsity loss to guide model focus foreground area, realize accurate positioning under no frame label;Multi-task joint optimization, through joint loss function completes model end-to-end fine-tuning.The application significantly improves the precision and anti-interference ability of insulator defect detection under zero sample / less sample working condition, realizes accurate positioning without manual boundary box labeling, reduces data labeling cost and landing threshold.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Power equipment identification and anomaly detection method and system based on cross-modal alignment, and storage medium

    CN121121425A

  • Bolt defect classification method and system based on feature decoupling and multi-modal alignment

    CN121414726A

  • Large-model-driven cross-modal power insulator defect detection method and system

    CN121437476A