A Method to Suppress False Alarms of Sentinel-1 to Improve Ship Detection KCI

Title
A Method to Suppress False Alarms of Sentinel-1 to Improve Ship Detection
Author(s)
Jeongju Bae; Yang, Chan Su
KIOST Author(s)
Yang, Chan Su(양찬수)
Alternative Author(s)
양찬수
Publication Year
2020-08
Abstract
In synthetic aperture radar (SAR) based ship detection application, false alarms frequently occur due to various noises caused by the radar imaging process. Among them, radio frequency interference (RFI) and azimuth smearing produce substantial false alarms; the latter also yields longer length estimation of ships than the true length. These two noises are prominent at cross-polarization and relatively weak at co-polarization.
However, in general, the cross-polarization data are suitable for ship detection, because the radar backscatter from background sea surface is much less in comparison with the co-polarization backscatter, i.e., higher shipsea image contrast. In order to improve the ship detection accuracy further, the RFI and azimuth smearing need to be mitigated. In the present letter, Sentinel-1 VV- and VH-polarization intensity data are used to show a novel technique of removing these noises. In this method, median image intensities of noises and background sea surface are calculated to yield arithmetic tendency. A band-math formula is then designed to replace the intensities of noise pixels in VH-polarization with adjusted VV-polarization intensity pixels that are less affected by the noises. To verify the proposed method, the adaptive threshold method (ATM) with a sliding window was used for ship detection, and the results showed that the 74.39% of RFI false alarms are removed and 92.27% false alarms of azimuth smearing are removed.
ISSN
1225-6161
URI
https://sciwatch.kiost.ac.kr/handle/2020.kiost/35175
DOI
10.7780/kjrs.2020.36.4.4
Bibliographic Citation
대한원격탐사학회지, v.36, no.4, pp.535 - 544, 2020
Publisher
대한원격탐사학회
Type
Article
Language
English
Files in This Item:
There are no files associated with this item.

qrcode

Items in ScienceWatch@KIOST are protected by copyright, with all rights reserved, unless otherwise indicated.

Browse