An image damage recognition method based on multi-scale feature fusion

By employing multi-scale feature fusion and a cascaded multi-task output structure, the problem of balancing local details and global semantic features in image damage recognition is solved, improving the accuracy and stability of damage recognition. This method is applicable to damage detection of turbine blades, industrial pipelines, and bridge structures.

CN121810684BActive Publication Date: 2026-05-26XIAN THERMAL POWER RES INST CO LTD +1
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
XIAN THERMAL POWER RES INST CO LTD
Filing Date
2026-03-09
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing technologies struggle to simultaneously consider both local detail features and global semantic features in image damage recognition. They also neglect the inherent dependencies between tasks during multi-task recognition and lack adaptive modeling and recognition stability for different damage types, especially exhibiting insufficient robustness under complex working conditions.

Method used

An image damage recognition method based on multi-scale feature fusion is adopted. By introducing channel attention mechanism and multi-head self-attention mechanism for feature extraction and enhancement, and combining bidirectional pyramid feature fusion network and cascaded multi-task output structure, the full fusion of multi-scale features and conditional dependency modeling between tasks are realized.

Benefits of technology

It improves the ability to identify damage of different sizes and shapes, and enhances the accuracy and stability of identification in complex backgrounds, especially significantly improving the detection accuracy in damage detection of turbine blades, industrial pipelines and bridge structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121810684B_ABST
    Figure CN121810684B_ABST
Patent Text Reader

Abstract

This invention provides an image damage recognition method based on multi-scale feature fusion, belonging to the fields of intelligent detection and computer vision technology. It can at least partially solve the problems of insufficient utilization of multi-scale features, negative transfer in multi-task recognition, and difficulty in adaptive modeling of different damage types in existing technologies. The invention includes: acquiring image data of the object to be detected or two-dimensional feature map data converted from one-dimensional signals and performing preprocessing; inputting the preprocessed data into a backbone feature extraction network with an attention mechanism to extract multi-level features; performing multi-head self-attention enhancement modeling on the high-level features; inputting the multi-level features and enhanced high-level features into a bidirectional pyramid feature fusion network for multi-scale fusion; and using a cascaded multi-task output structure based on the fused features to first predict the damage region and then predict the damage description. This invention achieves damage recognition under complex working conditions and effectively improves detection accuracy.
Need to check novelty before this filing date? Find Prior Art