Journal of Geo-information Science

MSF-UNet: A Mamba-based Spatial-Frequency Feature Fusion U-Net for Unsupervised Change Detection in SAR Images

ZHANG Yu, ZHUANG Huifu, ZHANG Xiang, TAN Zhixiang, LIU Yuhao, SHANG Jingjie, GUO Mingming

Journal of Geo-information Science, 2025, 27(9): 2213-2229. DOI: 10.12082/dqxxkx.2025.250269

方法类型	方法	漏检像元/个	虚警像元/个	总体误差/个	F1_Score/%	Kappa/%
传统	PCAKM	508	2 757	3 265	0.745	0.726
	INLPG	729	1 510	2 239	0.802	0.789
	IRG-McS	1 039	1 078	2 117	0.800	0.787
深度学习	DDNet	881	2 190	3 071	0.741	0.723
	FCD-GAN	41	4 843	4 884	0.682	0.655
	SNUNet-CD	1 736	84	1 820	0.795	0.785
	TransUNet	1 185	174	1 359	0.857	0.849
	CDRL-SA	1 129	477	1 606	0.838	0.828
	ELGC-Net	1 046	277	1 323	0.863	0.855
本文	MSF-UNet	921	94	1 015	0.896	0.890

Tab. 2 Quantitative analysis results of different methods on the Yellow River dataset

Other figure/table from this article

Fig. 1 The overall framework of the proposed method
Fig. 2 Flowchart of the DSCF method
Fig. 3 Structural diagram of the MSF-UNet model
Fig. 4 Images and reference change map of the Handan dataset
Fig. 5 Images and reference change map of the Yellow River dataset
Fig. 6 Impact analysis of threshold m on F1_Score of change detection results
Fig. 7 Change detection results of different methods on the Handan dataset
Tab. 1 Quantitative analysis results of different methods on the Handan dataset
Fig. 8 Change detection results of different methods on the Yellow River dataset
Fig. 9 Change detection results of the ablation experiment on the Handan dataset
Tab. 3 Evaluation metric results of ablation experiments on the Handan dataset
Tab. 4 Model parameter quantities and floating-point operation counts of different methods