Journal
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
Volume 59, Issue 6, Pages 5171-5182Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TGRS.2020.3010055
Keywords
Image segmentation; Semantics; Adaptation models; Data models; Feature extraction; Training; Buildings; Adversarial network; building extraction; domain adaptation; semantic segmentation
Categories
Funding
- Institute for Information & communications Technology Promotion (IITP) - Korea government (MSIP) (Geo-data generation and Applicable Service Development Based on Satellite Imagery Data Conversion Platform) [2018-0-01573]
- Basic Science Research program through the National Research Foundation of Korea (NRF) - MSIT [2020R1A2C1006295]
- Daegu Gyeongbuk Institute of Science and Technology (DGIST) Research and Development Program of the Ministry of Science and ICT [20-EE-01]
- National Research Foundation of Korea [2020R1A2C1006295] Funding Source: Korea Institute of Science & Technology Information (KISTI), National Science & Technology Information Service (NTIS)
Ask authors/readers for more resources
The article proposes a segmentation network based on a domain adaptive transfer attack (DATA) scheme for building extraction from aerial images, aiming to enhance the model's robustness and performance.
Semantic segmentation models based on convolutional neural networks (CNNs) have gained much attention in relation to remote sensing and have achieved remarkable performance for the extraction of buildings from high-resolution aerial images. However, the issue of limited generalization for unseen images remains. When there is a domain gap between the training and test data sets, the CNN-based segmentation models trained by a training data set fail to segment buildings for the test data set. In this article, we propose segmentation networks based on a domain adaptive transfer attack (DATA) scheme for building extraction from aerial images. The proposed system combines the domain transfer and the adversarial attack concepts. Based on the DATA scheme, the distribution of the input images can be shifted to that of the target images while turning images into adversarial examples against a target network. Defending adversarial examples adapted to the target domain can overcome the performance degradation due to the domain gap and increase the robustness of the segmentation model. Cross-data set experiments and ablation study are conducted for three different data sets: the Inria aerial image labeling data set, the Massachusetts building data set, and the WHU East Asia data set. Compared with the performance of the segmentation network without the DATA scheme, the proposed method shows improvements in the overall intersection over union (IoU). Moreover, it is verified that the proposed method outperforms even when compared with feature adaptation (FA) and output space adaptation (OSA).
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available