Railway substation ontology equipment integrity detection and defect positioning method and system based on deep learning multi-feature fusion

By using a deep learning-based multi-feature fusion method, visible light and infrared thermal images are simultaneously acquired and registered, enabling the integrity determination and precise defect location of railway substation equipment. This solves the problem of incomplete information in existing technologies, reduces labeling costs, and improves detection accuracy.

CN122391715APending Publication Date: 2026-07-14ZHONGXIN HANCHUANG BEIJING TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
ZHONGXIN HANCHUANG BEIJING TECH CO LTD
Filing Date
2026-04-17
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing methods for testing railway substation equipment suffer from incomplete single-modal information, high costs for obtaining finely labeled samples, and separation of integrity assessment and defect location, making it difficult to achieve both equipment integrity assessment and precise defect location.

Method used

A deep learning-based multi-feature fusion method is adopted to simultaneously acquire and register visible light images and infrared thermal images. Structural features and thermal distribution features are extracted through a dual-branch network. Combining multi-scale fusion and a two-layer cascade architecture, heterogeneous label collaborative training and progressive weight adjustment are used to achieve equipment integrity determination and precise defect location.

Benefits of technology

It improves the comprehensiveness and accuracy of equipment inspection, reduces the reliance on finely labeled samples, enables efficient integrity determination of equipment status and precise location of defects, improves the robustness of inspection and reduces labeling costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122391715A_ABST
    Figure CN122391715A_ABST
Patent Text Reader

Abstract

The application discloses a kind of railway substation ontology equipment integrity detection and defect positioning method and system based on deep learning multi-feature fusion.The method comprises: synchronously collecting and registering the visible light and infrared thermal image of equipment, obtaining two types of labels, and obtaining multi-modal image pair by augmentation;Input double-branch network to extract visible light structure features and infrared thermal distribution features to generate fusion discriminant feature map;Input double-layer cascade architecture in series, output pixel-by-pixel probability atlas and integrity binary classification result;Adopt heterogeneous label collaborative training, fuse two types of labels by using control switch parameters, gradually adjust the weight to balance two-level modules, use spatial distance modulation to differentiate the weight of defect pixels, and cut off the gradient back propagation from the second level to the first level to optimize the network parameters;When testing, output defect positioning result and integrity determination result and superimpose labels on original image.Realize equipment integrity determination and defect accurate positioning, reduce the dependence on fine annotation samples.
Need to check novelty before this filing date? Find Prior Art