| Title |
Shape Strengthened U-Shape Network For Objects Extraction Of Remote Sensing Images |
| ID_Doc |
48583 |
| Authors |
Xia Y.; Zhu B. |
| Year |
2022 |
| Published |
Proceedings - 2022 IEEE SmartWorld, Ubiquitous Intelligence and Computing, Autonomous and Trusted Vehicles, Scalable Computing and Communications, Digital Twin, Privacy Computing, Metaverse, SmartWorld/UIC/ATC/ScalCom/DigitalTwin/PriComp/Metaverse 2022 |
| DOI |
http://dx.doi.org/10.1109/SmartWorld-UIC-ATC-ScalCom-DigitalTwin-PriComp-Metaverse56740.2022.00183 |
| Abstract |
The developments of satellite and sensors technology has brought a significant increase in remote sensing (RS) images. It plays an important role to extract required objects from urban RS images automatically in smart cities. Semantic segmentation is a classic method to do that. In urban RS images, there are multi-scale objects with different shapes (e.g. buildings), which bring great challenges to the extraction. This paper proposes a shape strengthened U-shape network (SSU-Net), including a body stream and a shape stream. The body stream acquires the regular information of images, and a U-Net framework is designed to classify multi-scale objects. The shape stream helps to restore the lost shape information. Finally, two streams are fused by channel attention to focus on more discriminative information. Experimentally, the F1-score and IoU of the proposed SSU-Net reached 87.30% and 93.22% on WHU Buildings Dataset. And it achieves better performance compared with state-of-the-art methods. © 2022 IEEE. |
| Author Keywords |
Body stream; Objects extraction; Semantic segmentation; Shape stream; U-Net; Urban remote sensing |
