Cable damage detection method of yolo-cab model fusing multi-scale features and deformable convolution

By integrating multi-scale features with deformable convolution, the YOLO-CAB model solves the problems of high false negative rate for small targets, high computational complexity, and insufficient generalization ability in bridge cable damage detection, and achieves efficient and accurate damage detection.

CN122434834APending Publication Date: 2026-07-21YANGZHOU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
YANGZHOU UNIV
Filing Date
2026-04-03
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing bridge cable damage detection technologies suffer from problems such as high false negative rates for small targets, high computational complexity, insufficient generalization ability, difficulty in localization, and a lack of comprehensive ablation experiments and multi-model comparisons.

Method used

The YOLO-CAB model, which integrates multi-scale features and deformable convolution, is constructed by introducing an Extra Small Head for small target detection, Res-DSConv for residual depth separable convolution, and DCNv2 for deformable convolution. This improves detection accuracy and maintains real-time processing performance.

Benefits of technology

It significantly improves the ability to capture millimeter-level micro-damage, reduces computational complexity, enhances feature extraction efficiency, adapts to the nonlinear geometric deformation of cable damage, and provides strong algorithmic support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122434834A_ABST
    Figure CN122434834A_ABST
Patent Text Reader

Abstract

The application discloses a cable damage detection method of a YOLO-CAB model fusing multi-scale features and deformable convolution, S1 innovatively designs a special small target detection head, a Res-DSConv module and embeds deformable convolution DCNv2 in a prediction head, and adopts Focal-EIoU Loss to optimize the regression accuracy of a boundary box and an end-to-end training paradigm; S2 constructs a scene-rich bridge cable damage dataset, which contains annotated images of long-distance safe shooting, close-range fine shooting and various illumination conditions, and provides high-quality benchmark data for small target detection research; S3 compares the performance of mainstream models, carries out an ablation experiment, analyzes the parameter quantity and efficiency, and verifies actual cases, and comprehensively displays the advantages of YOLO-CAB in indexes and calculation efficiency, thereby providing a reliable reference benchmark for subsequent cable damage detection research. The YOLO-CAB network model provided by the application has generalization ability and robustness in a complex environment, and considers detection accuracy and real-time performance, and can provide reliable technical support for bridge automatic inspection.
Need to check novelty before this filing date? Find Prior Art