Remote Sensing Images based Road Network Extraction using Deep Learning: A Systematic Review

Abstract

Abstract Automatic extraction of roads based on data-driven methods is essential for various aspects such as route optimization, traffic management, GPS navigation, disaster management, defense, and security intelligence, etc. Due to the occlusion of trees, vehicles, buildings, etc., it is challenging. A systematic review is proposed in this study to overcome the challenges. More than 214 articles from 2018–2022 are collected using Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines from different databases (i.e., Scopus, Web of Science (WoS), IEEE Explorer), and 44 are reviewed. The article selection process is based on keywords like "Deep Learning (DL)," "Road Extraction," "High- Resolution Remote Sensing Images (HRRSI)," etc. The different datasets used by the researchers are also discussed in this study, along with the type of sensors and satellites used to collect HRRSI images. This study aims to provide a proposed solution to the investigations retrieved from the previous research work. After analysis, it is concluded that the factors retrieved from this rigorous analysis can be considered to propose a novel model that can resolve the issue of accurate extraction of roads.