Building extraction from high-resolution SAR imagery based on deep neural networks. Issue 9 (2nd September 2017)

• Record Type: Journal Article
• Authors: Xu, Zhen; Wang, Robert; Zhang, Heng; Li, Ning; Zhang, Lei
• Is Part Of: Remote sensing letters. Volume 8:Issue 9(2017)
• Page Start: 888
• Page End: 896
• Publication Date: 2017-09-02
• Subjects: Remote sensing -- Periodicals; Remote sensing; Periodicals; 621.3678
• DOI: 10.1080/2150704X.2017.1335906
• Languages: English
• ISSNs: 2150-704X

Abstract

ABSTRACT: Synthetic aperture radar (SAR) imagery classification is considered as one of the most significant SAR-based application. Developing SAR imagery classification applications based on new datasets requires considerable amount of work, such as feature extraction and validation. The classification processing flow is complex and should be specially designed for different SAR images. To reduce the complexity and improve the processing efficiency, a unified processing scheme based on the deep neural networks (DNN) is proposed. The scheme can be applied to most SAR imagery classification tasks by simply adjusting the model parameters. The proposed scheme is employed to extract building areas from different high resolution SAR images obtained by different sensors based on fully connected feedforward deep network (FDN) and convolutional neural network (CNN). The study results indicate that the proposed classification scheme has high accuracy and efficiency.