Image adaptive fusion method, device and terminal equipment

An adaptive fusion method that generates semantic and temperature-guided weights in intelligent inspection of industrial equipment solves the problems of unstable performance of key parts and high computational cost in existing technologies. It achieves efficient adaptive reconstruction of fused images, highlights temperature anomaly areas and preserves structural details.

CN122415348APending Publication Date: 2026-07-17CHINA TOBACCO GUANGXI IND +1

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHINA TOBACCO GUANGXI IND
Filing Date
2026-04-22
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing infrared and visible light image fusion methods have technical problems in intelligent inspection scenarios of industrial equipment, such as unstable performance of key parts, easy weakening of thermal anomalies, and high computational cost.

Method used

By inputting visible light and infrared images into the encoder for feature extraction, semantic guidance weights and temperature guidance weights are generated. Based on these weights, a joint guidance feature map is generated and adaptively fused. Finally, the fused image is reconstructed in the decoder.

Benefits of technology

This approach achieves the goal of highlighting temperature anomaly regions while preserving visible light structural details, resulting in richer detail representation of key components and reduced computational overhead.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415348A_ABST
    Figure CN122415348A_ABST
Patent Text Reader

Abstract

本申请涉及图像处理技术领域,公开了一种图像自适应融合方法、装置以及终端设备。所述方法包括:将可见光图像和红外图像输入至编码器,分别进行特征提取,得到红外特征和可见光特征;将可见光特征转换为每个部件区域对应的语义引导权重,以及将红外特征转换为每个部件区域对应的温度引导权重;基于语义引导权重和温度引导权重,生成联合引导特征图;将红外特征、可见光特征和联合引导特征图进行自适应融合,得到融合特征;将融合特征输入至解码器进行重建,得到自适应融合图像。采用上述方法可以使自适应融合图像在保留可见光结构细节的同时突出温度异常区域,使得关键部位的细节表现更加丰富。
Need to check novelty before this filing date? Find Prior Art