Relational Representation Learning Network for Cross-Spectral Image Patch Matching

Chuang Yu

doi:10.1016/j.inffus.2025.103749

Relational Representation Learning Network for Cross-Spectral Image Patch Matching

2025-09-15

0

PRE

AI

DOI：10.1016/j.inffus.2025.103749

Original

Original request for help

Save

Abstract

En 中文

• An innovative relational representation learning is proposed, which breaks the bottleneck of subsequent feature relation extraction caused by insufficient individual intrinsic feature mining in existing methods. • To further fully mine individual intrinsic features, a lightweight multi-dimensional global-to-local attention (MGLA) module is proposed and an efficient attention-based lightweight feature extraction (ALFE) network is built based on this module. • A multi-loss post-pruning (MLPP) optimization strategy is proposed that can promote network optimization without introducing additional parameters or increasing inference time. • Extensive experiments show that our RRL-Net achieves SOTA performance on multiple public cross-spectral datasets and has excellent robustness and generalization.

Relational Representation Learning Network for Cross-Spectral Image Patch Matching

Abstract

Journal

Organization