High-Resolution Remote Sensing Semantic Segmentation Method Coupling ResNet and Transformer

Method	IoU /%					mIoU /%	P /%	mF1 /%	FLOPs /10⁹	Params /10⁶
Method	Impervious surface	Building	Low vegetation	Tree	Car	mIoU /%	P /%	mF1 /%	FLOPs /10⁹	Params /10⁶
Swin-Unet^［19］	74.51	81.34	62.51	58.88	69.69	69.39	80.63	80.81	59.210	46.71
MAResU-Net^［14］	81.63	90.02	70.50	72.88	80.40	79.09	87.61	87.56	35.110	26.28
BANet^［22］	80.29	87.98	69.39	71.53	78.94	77.63	86.27	86.54	13.060	12.73
UNetFormer^［25］	80.87	89.58	69.55	71.99	79.16	78.23	86.55	86.95	11.740	11.68
TransAttUnet^［32］	79.71	86.87	68.77	70.51	78.57	76.89	80.34	77.14	355.310	22.65
CMTFNet^［10］	80.68	88.90	69.32	71.20	79.72	77.96	86.87	86.87	33.070	30.07
META-Unet^［33］	79.82	88.11	68.39	69.83	77.94	76.82	85.78	86.08	20.830	21.70
RTHNet （proposed）	81.54	90.16	71.23	73.13	81.84	79.58	87.94	87.91	33.880	31.08

Table 2. Comparison results of different methods on the Vaihingen dataset

View table

Table 2. Comparison results of different methods on the Vaihingen dataset

Method	IoU /%					mIoU /%	P /%	mF1 /%	FLOPs /10⁹	Params /10⁶
Method	Impervious surface	Building	Low vegetation	Tree	Car	mIoU /%	P /%	mF1 /%	FLOPs /10⁹	Params /10⁶
Swin-Unet^［19］	68.58	73.23	56.82	71.10	29.44	59.83	75.53	72.79	59.210	46.71
MAResU-Net^［14］	76.37	85.62	61.80	75.34	59.28	71.68	82.43	82.71	35.110	26.28
BANet^［22］	76.23	84.80	61.12	75.05	56.34	70.71	82.13	82.07	13.060	12.73
UNetFormer^［25］	76.25	85.37	62.64	76.11	62.74	72.62	83.78	82.83	11.740	11.68
TransAttUnet^［32］	74.38	81.60	61.05	73.94	50.10	68.21	80.74	80.12	355.310	22.65
CMTFNet^［10］	77.65	86.13	64.02	76.32	62.37	73.30	84.08	84.01	33.070	30.07
META-Unet^［33］	76.22	84.33	61.54	75.09	55.87	70.61	82.44	82.03	20.830	21.70
RTHNet （proposed）	78.43	86.79	64.30	75.80	62.74	73.61	84.03	84.20	33.880	31.08

Table 3. Comparison results of different methods on the WHDLD dataset

View table

Table 3. Comparison results of different methods on the WHDLD dataset

Method	IoU /%						mIoU / %	P / %	mF1 / %	FLOPs /10⁹	Params /10⁶
Method	Bare_soil	Buldding	Pavement	Road	Vegetation	Water	mIoU / %	P / %	mF1 / %	FLOPs /10⁹	Params /10⁶
Swin-Unet^［19］	31.50	53.95	37.82	55.13	78.83	89.15	57.73	72.61	69.82	7.760	27.15
MAResU-Net^［14］	30.02	55.21	38.82	60.26	78.60	89.46	58.73	73.78	70.59	8.780	26.23
BANet^［22］	32.32	55.29	39.17	59.15	78.67	89.60	59.03	73.15	71.01	3.260	12.73
UNetFormer^［25］	30.92	55.37	39.11	59.47	78.73	89.61	58.87	73.32	70.78	2.935	11.68
TransAttUnet^［32］	32.72	58.39	42.41	60.00	80.41	90.63	60.76	74.01	72.11	88.830	22.65
CMTFNet^［10］	33.20	57.49	40.54	61.23	79.39	90.20	60.34	74.14	72.16	8.570	30.07
META-Unet^［33］	31.89	54.79	38.84	59.31	78.17	89.46	58.74	73.04	70.81	5.230	21.70
RTHNet （proposed）	34.05	56.91	40.60	61.81	79.07	89.77	60.37	74.26	72.24	8.470	31.08

Table 4. Ablation experimental results on two datasets

View table

Table 4. Ablation experimental results on two datasets

Dataset	Baseline	AAFM	GLCTB	DEM	mIoU	mF1
Potsdam	√	×	×	×	77.42	86.51
	√	√	×	×	78.91	87.47
	√	√	√	×	79.30	87.83
	√	√	√	√	79.58	87.91
Vaihingen	√	×	×	×	69.24	80.86
	√	√	×	×	73.13	83.88
	√	√	√	×	73.27	84.00
	√	√	√	√	73.61	84.20

Tools

Get Citation

Copy Citation Text

Lei Zhang, Xue Ding, Jinliang Wang, Shuangyun Peng, Rongxiang Luo. High-Resolution Remote Sensing Semantic Segmentation Method Coupling ResNet and Transformer[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1628004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: Feb. 5, 2025

Accepted: Mar. 21, 2025

Published Online: Jul. 25, 2025

The Author Email: Xue Ding (4228@ynnu.edu.cn)

DOI:10.3788/LOP250591

CSTR:32186.14.LOP250591

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology