A novel target detection method for SAR images based on shadow proposal and saliency analysis

NEUROCOMPUTING

卷:267

Article

Gao, Fei¹  You, Jialing¹  Wang, Jun¹  Sun, Jinping¹  Yang, Erfu²  Zhou, Huiyu³ 

[1] Beihang Univ, Elect Informat Engn, Beijing 100191, Peoples R China

[2] Univ Strathclyde, Dept Design Mfg & Engn Management, Space Mechatron Syst Technol Lab, Glasgow G1 1XJ, Lanark, Scotland

[3] Queens Univ Belfast, Sch Elect Elect Engn & Comp Sci, Belfast BT7 1NN, Antrim, North Ireland

关键词: Synthetic aperture radar (SAR) target detection;    Shadow;    Saliency detection;    Local spatial autocorrelation;    Visual features;    One-Class SVM (OC-SVM);   

DOI  :  10.1016/j.neucom.2017.06.004

来源: Elsevier

【 摘 要 】

Conventional synthetic aperture radar (SAR) based target detection methods generally use high intensity pixels in the pre-screening stage while ignoring shadow information. Furthermore, they cannot accurately extract the target area and also have poor performance in cluttered environments. To solve this problem, a novel SAR target detection method which combines shadow proposal and saliency analysis is presented in this paper. The detection process is divided into shadow proposal, saliency detection and One-Class Support Vector Machine (OC-SVM) screening stages. In the shadow proposal stage, localizing targets is performed first with the detected shadow regions to generate proposal chips that may contain potential targets. Then saliency detection is conducted to extract salient regions of the proposal chips using local spatial autocorrelation and significance tests. Afterwards, in the last stage, the OC-SVM is employed to identify the real targets from the salient regions. Experimental results show that the proposed saliency detection method possesses higher detection accuracy than several state of the art methods on SAR images. Furthermore, the proposed SAR target detection method is demonstrated to be robust under different imaging environments. (C) 2017 Elsevier B.V. All rights reserved.

【 授权许可】

Free