Abstract
Automatic extraction of road information from remote sensing images is widely used in many fields, such as urban planning and automatic navigation. However, due to interference from noise and occlusion, the existing road extraction methods can easily lead to road discontinuity. To solve this problem, a road extraction network with bidirectional spatial information reasoning (BSIRNet) is proposed, in which neighbourhood feature fusion is used to capture spatial context dependencies and expand the receptive field, and an information processing unit with a recurrent neural network structure is used to capture channel dependencies. BSIRNet enhances the connectivity of road information through spatial information reasoning. Using the public Massachusetts road dataset and Wuhan University road dataset, the superiority of the proposed method is verified by comparing its results with those of other models.
Highlights
Roads play an important role in urban planning, traffic navigation, map updating, and other fields [1]
BSIRNet is based on DeepLabV3+ and consists of Xception, atrous spatial pyramid pooling (ASPP), spatial information inference structure (SIIS), spatial reasoning perception module (SRPM), and channel reasoning perception module (CRPM) components and a multiscale skip connection structure
The SRPM is used to capture the spatial context dependence, the CRPM is used to capture the dependence between channels, and the multiscale skip connection structure is used to capture more semantic information
Summary
Roads play an important role in urban planning, traffic navigation, map updating, and other fields [1]. With the rapid development of remote sensing satellites and sensors, it is becoming increasingly easy to collect very high-resolution (VHR) satellite imagery, which can provide sufficient data sources for road extraction. Extracting road information from VHR satellite imagery has become a popular topic of research. Researchers have developed many different road extraction methods [2], which can be generally divided into traditional methods and deep learning methods. Traditional road extraction methods rely on road image features and the construction of a theoretical model [3]. The image features used to extract roads in these methods are manually designed and lack an automatic learning process. Traditional road extraction methods have the disadvantages of low automation, complex operation, and high time consumption
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
