Topology-Aware Road Network Extraction via Multi-Supervised Generative Adversarial Networks

Abstract

Road network extraction from remote sensing images has played an important role in various areas. However, due to complex imaging conditions and terrain factors, such as occlusion and shades, it is very challenging to extract road networks with complete topology structures. In this paper, we propose a learning-based road network extraction framework via a Multi-supervised Generative Adversarial Network (MsGAN), which is jointly trained by the spectral and topology features of the road network. Such a design makes the network capable of learning how to “guess” the aberrant road cases, which is caused by occlusion and shadow, based on the relationship between the road region and centerline; thus, it is able to provide a road network with integrated topology. Additionally, we also present a sample quality measurement to efficiently generate a large number of training samples with a little human interaction. Through the experiments on images from various satellites and the comprehensive comparisons to state-of-the-art approaches on the public datasets, it is demonstrated that the proposed method is able to provide high-quality results, especially for the completeness of the road network.

An Ensemble Wasserstein Generative Adversarial Network Method for Road Extraction From High Resolution Remote Sensing Images in Rural Areas

“…[7] propose an improved GAN using the U-Net as G and suggest a simple loss function with an L2 loss and a cGAN loss. [18] create a multi-supervised GAN with two D to infer road networks with improved topology. However, these methods are too complex either in architecture (e.g.…”

“…two GANs in [15]) or loss function (e.g. four loss functions in [18]). Moreover, standard GANs struggle in vanishing gradients, mode collapse, and unstable training which are not easy to train [19,20].…”

“…It consists of a generator G and a discriminator D where G generates images that are indistinguishable from the ground truth and D distinguishes between the ground truth and the generated images. Few works have used standard GANs in road extraction from HR-RS images and achieved good performances [4,7,15,17,18]. However, these standard GANs may suffer from vanishing gradients, mode collapse, and unstable training [19,20].…”

VNet: An End-to-End Fully Convolutional Neural Network for Road Extraction From High-Resolution Remote Sensing Data

“…To the best of our knowledge, traditional techniques are time consuming and include numerous errors caused by human operators [12]. In recent years, researchers have proposed various kinds of approaches for road extraction from remote sensing images that include supervised [13] and unsupervised classification techniques [14]. These techniques generally utilize textural, photometric, and geometric characteristics to extract road parts, and they are based on image classification.…”

Estimating Road Segments Using Natural Point Correspondences of GPS Trajectories

“…Data-driven approaches have been developed to tackle the automated generation or to update existing road networks. A typical strategy is to use remote sensing data [1,2] or to gather information from traffic participants in the form of trajectories which then can be processed to obtain road segments. Trajectories are defined as a list of waypoints in metric space.…”