Exploring The Application of Deep Learning to Road Detection in Remote Sensing Images

Road information is a fundamental dataset in urban planning, traffic management, and related fields. However, challenges such as complex backgrounds, occlusions, and texture similarities with other linear features reduce the accuracy of road detection using remote sensing technology. This paper explores the advantages and recent developments in deep learning for enhancing road detection from remotely sensed images. The findings indicate that encoder-decoder semantic segmentation networks enable accurate road extraction; multi-scale feature fusion techniques enhance performance in occluded scenes; the integration of multi-task topology-constrained post-processing ensures road network connectivity and suppresses noise; and the adoption of weak supervision and lightweight models reduces labeling costs and improves deployment efficiency. Collectively, these advances will help remote sensing-based road detection reach new heights in terms of accuracy, intelligence, and breadth of application, driving the development of digital cities and intelligent transportation. This paper summarizes relevant techniques that support intelligent interpretation of remote sensing imagery and offers a valuable reference for future research in this field.