RoadSeg-CD: A Network With Connectivity Array and Direction Map for Road Extraction From SAR Images
Fei Gao,Jun Tu,Jun Wang,Amir Hussain,Huiyu Zhou
DOI: https://doi.org/10.1109/jstars.2022.3175594
IF: 4.715
2022-05-31
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Abstract:Road extraction from synthetic aperture radar (SAR) images has attracted much attention in the field of remote sensing image processing. General road extraction algorithms, affected by shadows of buildings and trees, are prone to producing fragmented road segments. To improve the accuracy and completeness of road extraction, we propose a neural network-based algorithm, which takes the connectivity and direction features of roads into consideration, named RoadSeg-CD. It consists of two branches: one is the main branch for road segmentation; the other is the auxiliary branch for learning road directions. In the main branch, a connectivity array is designed to utilize local contextual information and construct a connectivity loss based on the predicted probabilities of neighboring pixels. In the auxiliary branch, we proposed a novel road direction map, which is used for learning the directions of roads. The two branches are connected by specific feature fusion process, and the output from the main branch is taken as the road extraction result. Experiments on real radar images are implemented to validate the effectiveness of our method. The experimental results demonstrate that our method can obtain more continuous and more complete roads than several state-of-the-art road extraction algorithms.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geography, physical