Abstract
Road extraction is an important task in remote sensing image information extraction. Recently, deep learning semantic segmentation has become an important method of road extraction. Due to the impact of the loss of multiscale spatial features, the results of road extraction still contain incomplete or fractured results. In this article, we proposed a deep learning model, which is called the dense-global-residual network that reduces the loss of spatial information and enhances context awareness. In the dense-global-residual network, the residual network is used to extract the features at different levels. To obtain more abundant multiscale features, a dense and global spatial pyramid pooling module based on Atrous Spatial Pyramid Pooling is built to perceive and aggregate the contextual information. The proposed method obtains better results on the GF-2 road dataset and public Massachusetts road dataset of aerial imagery. In order to prove the effectiveness of our method, we compared with four methods, such as DeepLabV3+, U-net, D-LinkNet, and coord-dense-global model, and found that the accuracy of our method is considerably better. Moreover, the dense-global-residual network can also effectively extract roads, especially trees and building shadows that occlude the road. In addition, our method can successfully extract roads in regions of different development levels in universality experiments. This indicates that the proposed method can effectively maintain the completeness and continuity of roads and improve the accuracy of road segmentation from high-resolution remote sensing images.
Highlights
R OAD extraction from remote sensing images is an important topic in modern society
We analyze the problems of the fully convolutional network (FCN)-based methods, consider the road features, and construct an FCN network called the dense-global-residual network (DGRN) that is suitable for road extraction from high-resolution remote sensing images
The combination of the two improves the reuse of the road features, which reduces the loss of information, and aggregates multiscale contextual information to mitigate the effects of shadows, which improves the consistency of the semantic segmentation
Summary
R OAD extraction from remote sensing images is an important topic in modern society. It is of great significance to traffic management, urban planning, and map updating [1]–[3]. Feature-based methods have a good effect on simple and regular road extraction, but they have poor extraction effects on complex roads and require substantial postprocessing to repair the initially extracted roads. Classificationbased approaches such as the maximum likelihood methods (ML classification methods) [12], support vector machine methods (SVM classification methods) [13]–[15], Markov random fields classifier methods (MRF classification methods) [16]–[18], and mean shift-based methods [19] extract road fragments from the image and further refine them by customizing rules that are based on the spectral and spatial features of the road. Due to the spectral similarity of roads, buildings, parking lots, and other objects, the extraction accuracy is not high
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
