A Survey of Deep Learning Road Extraction Algorithms Using High-Resolution Remote Sensing Images

Shaoyi Mo,Yufeng Shi,Qi Yuan,Mingyue Li

DOI: https://doi.org/10.3390/s24051708

IF: 3.9

2024-03-07

Sensors

Abstract:Roads are the fundamental elements of transportation, connecting cities and rural areas, as well as people's lives and work. They play a significant role in various areas such as map updates, economic development, tourism, and disaster management. The automatic extraction of road features from high-resolution remote sensing images has always been a hot and challenging topic in the field of remote sensing, and deep learning network models are widely used to extract roads from remote sensing images in recent years. In light of this, this paper systematically reviews and summarizes the deep-learning-based techniques for automatic road extraction from high-resolution remote sensing images. It reviews the application of deep learning network models in road extraction tasks and classifies these models into fully supervised learning, semi-supervised learning, and weakly supervised learning based on their use of labels. Finally, a summary and outlook of the current development of deep learning techniques in road extraction are provided.

engineering, electrical & electronic,chemistry, analytical,instruments & instrumentation

What problem does this paper attempt to address?

### Problems the paper attempts to solve

This paper aims to systematically review and summarize the road extraction algorithms for high - resolution remote sensing images based on deep learning. Specifically, the paper focuses on the following aspects:

1. **Importance of road extraction**:
- As a basic element of transportation, roads connect urban and rural areas and influence people's daily lives and work. Road extraction plays an important role in many fields such as map updating, economic development, tourism and disaster management.

2. **Existing challenges**:
- The extraction of road information from high - resolution remote sensing images has always been a hot and difficult problem. These challenges include complex backgrounds (such as buildings, vegetation, various types of roads), diverse road shapes (different widths and lengths) and poor image perspectives (cloud and fog occlusion, lighting effects, etc.). With the expansion of cities, the topological structure of roads has become extremely complex, and many buildings have blocked large areas of road areas.

3. **Application of deep learning**:
- Deep learning network models have been widely used to extract roads from remote sensing images in recent years. The paper classifies these models into fully - supervised learning, semi - supervised learning and weakly - supervised learning according to the label usage.

4. **Classification of methods**:
- Fully - supervised learning depends on a large number of pixel - level training labels. Although it can achieve a high - precision segmentation structure, its generalization ability is weak, and obtaining pixel - level labels requires a large amount of manual annotation work.
- Semi - supervised (weak) learning depends on less training label data, which can be weak labels in the form of points, lines, etc. Although it is slightly inferior to fully - supervised learning in segmentation performance, it reduces the dependence on label data and alleviates the burden of manual annotation.

5. **Future prospects**:
- The paper also discusses the limitations of current deep - learning technologies in road extraction and proposes future development directions and challenges.

### Summary

By systematically reviewing and summarizing the road extraction algorithms based on deep learning, this paper aims to solve the challenges of road information extraction in high - resolution remote sensing images, improve the accuracy and efficiency of road extraction, and provide references and guidance for future research.