Dense Hybrid Attention Network for Remote Sensing Building Change Detection

Qinglin Tian; Donghua Lu; Yao Li; Chengkai Pei

doi:10.3788/AOS241436

Journals >Acta Optica Sinica >Volume 45 >Issue 6 >Page 0628008 > Article

Acta Optica Sinica
Vol. 45, Issue 6, 0628008 (2025)

Dense Hybrid Attention Network for Remote Sensing Building Change Detection

Qinglin Tian^1,*, Donghua Lu¹, Yao Li², and Chengkai Pei¹

Author Affiliations

¹National Key Laboratory of Uranium Resources Exploration-Mining and Nuclear Remote Sensing, Beijing Research Institute of Uranium Geology, Beijing 100029, China

²School of Geographical Sciences, Southwest University, Chongqing 400715, China

show less

DOI: 10.3788/AOS241436 Cite this Article Set citation alerts

Qinglin Tian, Donghua Lu, Yao Li, Chengkai Pei. Dense Hybrid Attention Network for Remote Sensing Building Change Detection[J]. Acta Optica Sinica, 2025, 45(6): 0628008 Copy Citation Text

show less

Fig. 1. Overview architecture of proposed DHANet

Download full size

Fig. 2. Convolutional block attention module

Download full size

Fig. 3. Diagrams of traditional convolution and dilated convolution. (a) Traditional convolution; (b) dilated convolution

Download full size

Fig. 4. Multi-scale features aggregation module

Download full size

Fig. 5. Hybrid attention module

Download full size

Fig. 6. Interlaced sparse self-attention module

Download full size

Fig. 7. Data sets. (a) LEVIR-CD; (b) WHU-CD

Download full size

Fig. 8. Qualitative results on LEVIR-CD dataset

Download full size

Fig. 9. Qualitative results on WHU-CD dataset

Download full size

Method	P /%	R /%	F₁ /%	IoU /%
FC-EF	86.91	80.17	83.40	71.53
FC-Siam-Conc	91.99	76.77	83.69	71.96
FC-Siam-Diff	89.53	83.31	86.31	75.92
STANet	83.81	91.00	87.26	77.40
IFN	93.02	86.33	89.55	81.08
BIT	91.89	88.48	90.15	82.07
DHANet	92.02	90.08	91.04	83.55

Table 1. Quantitative results on LEVIR-CD dataset

Method	P /%	R /%	F₁ /%	IoU /%
FC-EF	86.44	71.24	78.11	64.08
FC-Siam-Conc	87.55	75.10	80.85	67.85
FC-Siam-Diff	86.40	76.64	81.23	68.39
STANet	75.70	89.85	82.17	69.74
IFN	91.65	82.94	87.08	77.11
BIT	89.65	85.04	87.28	77.44
DHANet	90.86	87.91	89.36	80.77

Table 2. Quantitative results on WHU-CD dataset

DC	MSA	HAM	P /%	R /%	F₁ /%	IoU /%
×	×	×	89.83	87.64	88.72	79.73
√	×	√	90.62	89.09	89.85	81.57
√	√	×	90.55	89.25	89.90	81.65
×	√	√	91.83	89.37	90.58	82.79
√	√	√	92.02	90.08	91.04	83.55

Table 3. Quantitative results of ablation experiments

Method	Params /M	FLOPs /G	F₁ /%
FC-EF	1.35	2.92	83.40
FC-Siam-Conc	1.54	4.55	83.69
FC-Siam-Diff	1.35	3.99	86.31
STANet	23.32	23.84	87.26
IFN	50.71	82.26	89.55
BIT	3.55	10.60	90.15
DHANet	21.65	32.26	91.04

Table 4. Comparisons of complexity on LEVIR-CD dataset

Qinglin Tian, Donghua Lu, Yao Li, Chengkai Pei. Dense Hybrid Attention Network for Remote Sensing Building Change Detection[J]. Acta Optica Sinica, 2025, 45(6): 0628008

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享