A Comparative Study of Deep Learning Methods for Automated Road Network Extraction from High-Spatial-Resolution Remotely Sensed Imagery
Bibliographic record
Abstract
Road network data are crucial for various applications, such as road network planning, traffic control, map navigation, autonomous driving, and smart city construction. Automated road network extraction from high-spatial-resolution remotely sensed imagery has shown promise in road network data construction. In recent years, the advent of deep learning algorithms has pushed road network extraction towards auto - mation, achieving very high accuracy. However, the latest deep learning models are often less applied in the field of road network extraction and lack comparative experiments for guidance. Therefore, this research selected three recent deep learning algorithms, including dense prediction transformer (DPT), SegFormer, SEgmentation TRansformer (SETR), and the classic model fully convolutional network-8s (FCN-8s) for a comparative study. Additionally, this research paper compares three different decoder structures within the SETR model (SETR_naive, SETR_mla, SETR_pup) to investigate the effect of different decoders on the road network extraction task. The experiment is conducted on three commonly used datasets: the DeepGlobe Dataset, the Massachusetts Dataset, and Road Datasets in Complex Mountain Environments (RDCME). The DPT model outperforms other models on the Massachusetts dataset with superior reliability, achieving a high accuracy of 96.31% and excelling with a precision of 81.78% and recall of 32.50%, leading to an F1 score of 46.51%. While SegFormer has a slightly higher F1 score, DPT's precision is particularly valuable for minimizing false positives, making it the most balanced and reliable choice. Similarly, for the DeepGlobe Dataset, DPT achieves an accuracy of 96.76%, precision of 66.12%, recall of 41.37%, and F1 score of 50.89%, and for RDCME, DPT achieves an accuracy of 98.94%, precision of 99.07%, recall of 99.84%, and F1 score of 99.46%, confirming its consistent performance across datasets. This paper provides valuable guidance for future studies on road network extraction techniques using deep learning algorithms.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.009 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.004 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.004 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".