Adaptive Dual-Stream Enhanced Network for Heterogeneous Remote Sensing Change Detection

Yaxi Chen1

Cheng Qian1

Chang Liu2

Donglai Tang3

Liang Liu4, Email

Xianmin Mai5, Email

1College of Computer Science and Artificial Intelligence, Southwest Minzu University, Chengdu, Sichuan, 610000, China
2College of Computer Science, Chengdu University, Chengdu, Sichuan, 610106, China
3Aostar Information Technology Co., Ltd, Chengdu, Sichuan, 610299, China
4College of Architecture and Civil Engineering, Chengdu University, Chengdu, Sichuan, 610106, China
5College of Architecture, Southwest Minzu University, Chengdu, Sichuan, 610225, China

 

Abstract

Heterogeneous remote sensing change detection (HRSCD) is pivotal for disaster emergency monitoring and national geographic surveying. However, the inherent discrepancies in imaging mechanisms between optical and synthetic aperture radar (SAR) sensors often lead to pseudo-changes and blurred semantic boundaries in existing methods. To address these issues, this paper presents an Adaptive Dual-stream Enhanced Network (ADEN). First, a weight-sharing feature extraction module is employed to extract hierarchical multi-scale features from bi-temporal heterogeneous images. Subsequently, to bridge the modality gap, a Hierarchical Alignment Module (HAM) is designed to progressively fuse multi-dimensional information, achieving layer-wise alignment from shallow feature space to semantic space. Finally, a Multi-scale Feature Decoder (MFD) performs multi-level feature decoding to precisely identify change regions. Experiments conducted on five benchmark heterogeneous datasets demonstrate that ADEN effectively suppresses pseudo-changes and yields sharper boundary identification. The proposed method achieves a mean Intersection over Union (mIoU) of 83.32%, which is 3.5% higher than the second-best approach. This study provides a solution for HRSCD. The source code and datasets are available at https://github.com/1234qiancheng1234/ADEN.

Adaptive Dual-Stream Enhanced Network for Heterogeneous Remote Sensing Change Detection