Road Information Extraction and Application Based on Multi-Modal Data

This paper introduces an automated method for road information extraction utilizing neural networks. It enhances the precision and connectivity of road segmentation in remote sensing imagery by capitalizing on the distinct features of road centerlines and widths. We introduce the Average Path Length Similarity (APLS) metric to assess connectivity and develop an optimization algorithm for road centerlines, transforming segmentation predictions into accurate vectorized representations. The method's efficacy is validated through experiments, showcasing its ability to extract continuous road networks accurately across various datasets, thereby offering substantial technical support for urban planning and transportation management applications.