| 講演抄録/キーワード |
| 講演名 |
2022-03-10 09:00
Deep Cascade Road Extraction Network: a Multi-task Method for Road Extraction ○Yubo Wang(Waseda Univ.)・Zhao Wang・Yuusuke Nakano・Ken Nishimatsu(NTT)・Katsuya Hasegawa(JAXA)・Jun Ohya(Waseda Univ.) PRMU2021-60 |
| 抄録 |
(和) |
In this work, we present an end-to-end cascade neural network model called Deep Cascade Road Extraction Network to extract accurate road networks from aerial imagery. On the basis of the cascade structure consisting of three subnetworks (Surface Segmentation Network, Edge Detection Network, and Centreline Extraction Network) connected in a cascade manner, we simultaneously achieve the three tasks of the subnetworks for road extraction. Through comparison experiments, our method achieves state-of-the-art results for all the three subtasks. Meanwhile, our model demonstrates strong robustness to occlusions while accurately extracting complex road areas. |
| (英) |
In this work, we present an end-to-end cascade neural network model called Deep Cascade Road Extraction Network to extract accurate road networks from aerial imagery. On the basis of the cascade structure consisting of three subnetworks (Surface Segmentation Network, Edge Detection Network, and Centreline Extraction Network) connected in a cascade manner, we simultaneously achieve the three tasks of the subnetworks for road extraction. Through comparison experiments, our method achieves state-of-the-art results for all the three subtasks. Meanwhile, our model demonstrates strong robustness to occlusions while accurately extracting complex road areas. |
| キーワード |
(和) |
Aerial Image Segmentation / Road Extraction / Multi-task Model / / / / / |
| (英) |
Aerial Image Segmentation / Road Extraction / Multi-task Model / / / / / |
| 文献情報 |
信学技報, vol. 121, no. 427, PRMU2021-60, pp. 1-6, 2022年3月. |
| 資料番号 |
PRMU2021-60 |
| 発行日 |
2022-03-03 (PRMU) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
PRMU2021-60 |