Article: Improved DeepLabv3+ connected augmented reality technology for building target extraction in urban environmental design Journal: International Journal of Information and Communication Technology (IJICT) 2024 Vol.24 No.5 pp.54 - 73 Abstract: Aiming at the problem of inaccurate segmentation of building edges in remote sensing images, the imprecise segmentation of building edges in remote sensing images by deep learning models is an important research direction for remote sensing intelligence applications. This paper proposes a lightweight remote sensing image building extraction method based on DeepLabv3. The skeleton network uses DeepLabv3 to connect the IEU-Net structure. Secondly, in order to solve the problem of limited feature richness of the model, the morphological construction index MBI is introduced to participate in the classification process of the model together with the RGB band of the remote sensing image. Finally, in the model prediction, corresponding to IELoss, a strategy of ignoring edge prediction is adopted to obtain the best building extraction results. Our proposed method can effectively overcome the problem of insufficient edge pixel features of samples, suppress the influence of road and building shadows on the results, and improve the extraction accuracy of houses and buildings in remote sensing images. Inderscience Publishers - linking academia, business and industry through research

Title: Improved DeepLabv3+ connected augmented reality technology for building target extraction in urban environmental design

Authors: Jie Chen; Qian Wu

Addresses: Sichuan Technology and Business University, Chengdu 610000, Sichuan, China ' Sichuan Technology and Business University, Chengdu 610000, Sichuan, China

Abstract: Aiming at the problem of inaccurate segmentation of building edges in remote sensing images, the imprecise segmentation of building edges in remote sensing images by deep learning models is an important research direction for remote sensing intelligence applications. This paper proposes a lightweight remote sensing image building extraction method based on DeepLabv3. The skeleton network uses DeepLabv3 to connect the IEU-Net structure. Secondly, in order to solve the problem of limited feature richness of the model, the morphological construction index MBI is introduced to participate in the classification process of the model together with the RGB band of the remote sensing image. Finally, in the model prediction, corresponding to IELoss, a strategy of ignoring edge prediction is adopted to obtain the best building extraction results. Our proposed method can effectively overcome the problem of insufficient edge pixel features of samples, suppress the influence of road and building shadows on the results, and improve the extraction accuracy of houses and buildings in remote sensing images.

Keywords: building extraction; boundary perception; DeepLabv3+.

DOI: 10.1504/IJICT.2024.138438

International Journal of Information and Communication Technology, 2024 Vol.24 No.5, pp.54 - 73

Received: 15 Apr 2023
Accepted: 15 Nov 2023
Published online: 03 May 2024 *

Title: Improved DeepLabv3+ connected augmented reality technology for building target extraction in urban environmental design

Keep up-to-date