| Creator |
|
|
|
| Language |
|
| Publisher |
|
|
|
| Date of Available |
|
| Date |
|
| Source Title |
|
| Vol |
|
| Issue |
|
| First Page |
|
| Last Page |
|
| Publication Type |
|
| Access Rights |
|
| Crossref DOI |
|
| Rights |
|
| Abstract |
This work aims to examine the function of deep learning (DL) in improving building facade segmentation to facilitate sustainable urban planning and green building (GB) assessment. The study used a lar...ge collection of open-source façade images with 51,731 architectural elements labeled. To address class imbalance, six methods for data augmentation were used. The study utilized a progressive model augmentation technique. First, the original dataset was used to train basic U-Net architecture. Second, Canny Edge Detection (CED) was included to improve the presentation of boundaries and structures. Third, an attention mechanism was added to improve feature selection and contextual learning. Lastly, two improvements were made to the architecture: a learnable edge branch for adaptive boundary modeling and a boundary-aware hybrid loss function to improve contour accuracy. Experimental results show that the proposed new framework performs better at segmentation, achieving an overall accuracy of 0.982 and improved border consistency. The combination of edge-guided learning and boundary optimization yields much better facade delineation than a regular U-Net-based model. Proposed identifying architectural parts makes facade assessment more reliable, which supports energy- efficient design analysis, sustainable retrofitting plans, and environmentally friendly urban development. The suggested approach provides a robust computational resource for architects, urban planners, and sustainability researchers assessing green infrastructure.show more
|
Export Mendeley
p141-155