A steel surface defect recognition method based on state space model feature fusion and texture enhancement

By improving the MobileMamba state space model and multi-scale adaptive feature fusion module, and combining it with a defect-sensitive texture enhanced attention module, the problems of limited receptive field and high computational complexity in existing strip steel surface defect classification methods are solved, achieving high-precision real-time detection and quality control.

CN122415486APending Publication Date: 2026-07-17FUJIAN UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
FUJIAN UNIV OF TECH
Filing Date
2026-04-16
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing methods for classifying surface defects in strip steel based on CNNs such as ResNet suffer from problems such as limited receptive field, difficulty in modeling long-range spatial dependence, unstable accuracy of multi-scale defect classification, insufficient ability to distinguish visually similar defects, and high computational complexity of Transformer-type methods, which makes it difficult to meet the requirements of real-time detection.

Method used

An improved MobileMamba state-space model is used as the backbone network. Feature extraction is performed through a three-stage hierarchical architecture. Combined with a multi-scale adaptive feature fusion module and a defect-sensitive texture enhancement attention module, global context modeling and multi-scale feature fusion are achieved, background noise and illumination artifacts are suppressed, and classification accuracy is improved.

Benefits of technology

While maintaining lightweight computing overhead, it significantly improves the classification accuracy of surface defects in strip steel, making it suitable for real-time automatic detection and quality control on industrial production lines.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415486A_ABST
    Figure CN122415486A_ABST
Patent Text Reader

Abstract

The application provides a steel surface defect recognition method based on state space model feature fusion and texture enhancement, comprising: inputting a preprocessed image into a trained defect recognition model to obtain a defect classification result; wherein the defect recognition model comprises a backbone network, a multi-scale adaptive feature fusion module and a defect-sensitive texture enhancement attention module connected in sequence; the backbone network is a three-stage hierarchical feature extraction network constructed based on a selective state space model, each feature extraction block contains a global branch, a local branch and an identity branch; the multi-scale adaptive feature fusion module is used for channel dimension alignment and spatial size alignment of the feature map output by the backbone network, and outputs a fusion feature map; the defect-sensitive texture enhancement attention module is used for multi-branch parallel multi-scale texture feature extraction on the fusion feature map, and outputs a defect enhancement feature map; and the defect class of the steel surface image is determined according to the defect enhancement feature map.
Need to check novelty before this filing date? Find Prior Art