MADDRNet: A Morphology-Aware Dual-Domain Resonance Network for Remote Sensing Change Detection
Hanming Wei, Yujiang He, Lu Bai, Yuanxu Wang, Xin Yang, Junyi Lv, Yurong Qian, Weijun Gong
Xinjiang University Hexi University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Remote sensing change detection (RSCD) under environmental interference, such as climatic variations and heterogeneous landforms, remains a challenging problem. Environmental variations often induce spectral style drift, which makes it difficult for existing methods to distinguish true semantic changes from pseudochanges. Meanwhile, anisotropic targets (e.g., winding rivers or fragmented farmland) pose additional challenges for morphology preservation, since rigid convolutional sampling tends to produce discontinuous structural representations. To address the representation dilemma between semantic generalization and fine-grained morphology preservation, we propose a morphology-aware dual-domain resonance network (MADDRNet), a novel semantic–morphology collaborative framework. Unlike conventional Siamese feature extraction architectures, the proposed framework integrates a frozen vision foundation model (VFM) to construct an environment-invariant macrolevel semantic prior, while a learnable branch focuses on capturing fine-grained structural details. To mitigate spectral style drift caused by environmental variations, a dual-domain resonance module (DDRM) is introduced to decouple semantic representations from climate-induced appearance discrepancies by integrating frequency-domain spectral gating with spatial-domain edge-aware feature injection. Furthermore, a morphology-aware gated fusion (MAGF) module dynamically adapts its receptive field to irregular target boundaries while suppressing nonsemantic background noise through temporal interaction. Extensive experiments on four benchmark datasets—CLCD, GFSWCLCD, LuojiaSET-CLCD, and Water-CD—demonstrate that the proposed method achieves state-of-the-art performance under diverse environmental conditions. This work provides a promising solution for robust RSCD under complex environmental interference and offers a new perspective for jointly modeling semantic generalization and fine-grained morphology preservation.
逐年被引趋势
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIRemote-Sensing Image Classification
Remote Sensing and Land Use · Remote Sensing in Agriculture