Robust Infrared Image Feature Modeling and Matching for Power Distribution Line Inspection

Infrared imaging has gained increasing traction in power distribution line inspection due to its effectiveness in enhancing operational safety. Infrared image matching serves as a key technology for re-identifying overheated components and supports upstream tasks such as UAV navigation and 3D reconstruction. However, most existing image matching algorithms are designed for visible light images and fail to address the unique characteristics of infrared imagery, including weak textures and low resolution. To overcome these challenges, we propose RI-FMM (Robust Infrared Feature Modeling and Matching), a detector-free matching framework specifically tailored for infrared image data. RI-FMM introduces the simplified bilinear interpolation attention module (SBAM), which incorporates a pixel-level attention mechanism, into the feature modeling stage prior to matching. This approach enhances the network’s ability to extract discriminative infrared features while reducing computational overhead. Extensive experiments on real infrared images from power distribution lines demonstrate that RI-FMM achieves significantly improved matching performance compared to existing methods, offering a robust solution for infrared-based inspection tasks.