Multi‐task fully convolutional networks for building segmentation on SAR image. Issue 20 (18th September 2019)

• Record Type: Journal Article
• Authors: Zhang, Zenghui; Guo, Weiwei; Yu, Wenhao; Yu, Wenxian
• Is Part Of: Journal of engineering. Volume 2019:Issue 20(2019)
• Page Start: 7074
• Page End: 7077
• Publication Date: 2019-09-18
• Subjects: radar imaging -- image classification -- image segmentation -- synthetic aperture radar; branched sub‐task -- entire networks -- main network -- related sub‐task -- particular buildings targets -- multitask fully convolutional networks -- building segmentation -- SAR image -- optical image semantic segmentation -- similar performance gain -- synthetic aperture radar image -- missing alarm rate -- building targets -- multitask FCN -- original FCN; Engineering -- Periodicals; Engineering; Electronic journals; Periodicals; 620.005
• DOI: 10.1049/joe.2019.0569
• Languages: English
• ISSNs: 2051-3305

Abstract

Abstract : Fully convolutional networks (FCN) and related advanced variants have attached outstanding performance on optical image semantic segmentation. However, they do not obtain similar performance gain on synthetic aperture radar (SAR) image if they are directly applied due to the differences between optical and SAR images. The missing alarm rate of FCN remains high for some building targets. In this study, a multi‐task FCN is proposed for building extraction from SAR images. The main task remains the same as the original FCN, and in addition, a branched sub‐task is designed to extract the pivotal parts of buildings to make the entire networks pay more attention to the high backscattering intensity parts of buildings. The main network then can be boosted by the related sub‐task and reduce the missing rate. The experimental results on the same data set show that this model reaches low missing rate and demonstrate the efficiency especially for some particular buildings targets.