MDCNet: A Multiscale Dual-Domain Collaborative Network for Semantic Segmentation of Remote Sensing Images

Abstract

Semantic segmentation is a fundamental task in remote sensing image (RSI) analysis and plays a vital role in applications such as land-use mapping, urban planning, and environmental monitoring. However, RSIs often suffer from severe grayscale variations and unstable interclass variance, which lead to blurred category boundaries and reduced segmentation accuracy. Most existing methods primarily focus on spatial-domain feature modeling; however, they pay insufficient attention to the rich information embedded in the frequency domain, which consequently limits segmentation performance in complex scenarios. To address these challenges, this article proposes a multiscale dual-domain collaborative network (MDCNet), which innovatively introduces a spatial-frequency collaborative module (SFCM) to separate high- and low-frequency information using discrete wavelet transform (DWT) and to deeply fuse spatial- and frequency-domain features through a cross-attention mechanism. In addition, a multiscale gated perception bridge (MGB) is designed to promote efficient interaction among multiscale features, and a multidimensional feature refinement head (MFRH) is introduced to reduce the semantic gap between deep and shallow features, thereby improving overall segmentation performance. Extensive experiments on the ISPRS Vaihingen, ISPRS Potsdam, and LoveDA datasets demonstrate that MDCNet consistently outperforms existing methods in segmentation accuracy and robustness.