An Enhanced Trans-Involution Network for Building Footprint Extraction from High Resolution Orthoimagery
Bibliographic record
Abstract
The amalgamation of data visualization and geospatial insights has driven significantly advancements in remote sensing for applications like damage detection and urban planning, particularly building rooftop extraction from high spatial satellite imagery. However, building rooftop extraction using deep learning methods often results in outputs with unclear margin delineation. In this study, we propose a novel approach that combines Transformer architectures, involution, and an enhanced U-net (E-Unet) [1] to improve building footprint extraction performance. Our method demonstrates remarkable accuracy in complex urban environments in the Waterloo Building Dataset. Transformers, renowned for their success in natural language processing, have excelled in adeptness at analyzing sequential data. By using the embedding and multi-head attention blocks, this method is becoming increasingly valuable for building extraction. Involution, in turn, augments neural networks by providing spatial-specific adaptability, effectively extracting inter-band features and surpassing convolutional limitations. Through comprehensive comparative model experiments on the Waterloo building dataset, the optimal architecture was identified. The model significantly enhances accuracy when the Transformer architecture is integrated at the output of the E-Unet. Our proposed network achieves outstanding at the crucial metric values of IoU, mIoU, Precision, F1-score in 81.2, 91.9, 92.9, 89.8 (%), surpassing established frameworks such as FCN-8s, U-Net, DeepLab v3+, Fast Statistical Convolutional Neural Network (SCNN), High-Resolution Net (HRNet) v2, Mask R-CNN, as well as E-Unet.
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 | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| 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".