Researching | Remote Sensing Road Extraction Combining Contextual Information and Multi-Layer Features Fusion

Laser & Optoelectronics Progress, Volume. 61, Issue 4, 0428007(2024)

Remote Sensing Road Extraction Combining Contextual Information and Multi-Layer Features Fusion

Guo Chen^1,2 and Likun Hu^1,2、*

Author Affiliations

¹School of Electrical Engineering, Guangxi University, Nanning 530004, Guangxi, China

²Advanced Measurement & Control & Intelligent Power Research Center, Guangxi University, Nanning 530004, Guangxi, China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (15)

References (28)

Paper Information

Figures & Tables(15)

Fig. 1. Structure of CMF-UNet

Download full size

Fig. 2. Pyramid feature aggregation module

Download full size

Fig. 3. Channel attention module

Download full size

Fig. 4. Spatial attention module

Download full size

Fig. 5. CBAM

Download full size

Fig. 6. Dilated convolutiona in D-LinkNet

Download full size

Fig. 7. Multis-scale context extraction module

Download full size

Fig. 8. Massachusetts Roads dataset

Download full size

Fig. 9. CHN6-CUG dataset

Download full size

Fig. 10. Loss curves of the CMF-UNet training process

Download full size

Fig. 11. Segmentation results on Massachusetts Roads

Download full size

Fig. 12. Segmentation results on CHN6-CUG

Download full size

Table 1. Ablation study

Table 1. Ablation study

Base	PFA	MSCE	Dataset	R_recall /%	s_F1 /%	R_IoU /%
√			MassachusettsRoads	69.83	75.59	60.79
√			CHN6-CUG	72.09	77.73	63.57
√	√		MassachusettsRoads	70.71	76.29	61.67
√	√		CHN6-CUG	75.18	78.56	64.7
√		√	MassachusettsRoads	72.13	76.56	62.08
√		√	CHN6-CUG	74.68	78.8	65.02
√	√	√	MassachusettsRoads	72.52	76.87	62.29
√	√	√	CHN6-CUG	76.27	79.11	65.45

Table 2. Comparison of algorithm complexity
View table
Table 2. Comparison of algorithm complexity
Model FLOPs /10⁹ Parameter quantity /10⁶
PSPNet 118.45 46.72
D-LinkNet 67.33 31.09
DeepLabv3+ 166.84 54.71
U-Net 451.67 24.89
CMF-UNet 451.73 24.93

Table 3. Test results on different datasets

Table 3. Test results on different datasets

Model		Massachusetts Roads				CHN6-CUG
Model		R_recall /%	s_F1 /%	R_IoU /%		R_recall /%	s_F1 /%	R_IoU /%
PSPNet		30.09	39.95	24.96		30.58	42.72	27.17
U-Net		69.83	75.59	60.79		72.09	77.73	63.57
DeepLabv3+		68.73	66.37	49.66		70.07	68.64	52.26
D-LinkNet		70.86	75.04	60.52		70.63	70.39	58.74
CMF-UNet		75.60	77.61	63.41		78.56	79.26	65.61

Tools

Get Citation

Copy Citation Text

Guo Chen, Likun Hu. Remote Sensing Road Extraction Combining Contextual Information and Multi-Layer Features Fusion[J]. Laser & Optoelectronics Progress, 2024, 61(4): 0428007

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: Apr. 3, 2023

Accepted: Jun. 20, 2023

Published Online: Feb. 26, 2024

The Author Email: Likun Hu (hlk3email@163.com)

DOI:10.3788/LOP231024

CSTR:32186.14.LOP231024

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology