Removal of Different Types of Noises in Synthetic Aperture Radar (SAR) Images for Improved Ship Detection

Abstract

Synthetic Aperture Radar (SAR) images contain different types of noises which restricts its wide application for the ocean surveillance. Therefore, this study focuses on removing several types of noises from SAR images. At first images were Fourier transformed to obtain frequency domain. Then, sidelobe noises from KOMPSAT-5, and scalloping and thermal noises from Sentinel-1 images were masked out by applying low-pass filter on the frequency domain. Then pixels affected by azimuth ambiguity in KOMPSAT-5 images were determined based on the distance and comparative brightness of the detected ships, and removed accordingly. This method is applied on 4 KOMPSAT-5 images and validated with the visual detection results of ships. The ship detection results without applying noise removal contains up to 59.26% false detections which were fully removed by the proposed method. Thus, the proposed noise reduction scheme has improved the accuracy of ship detection. Further improvement of the algorithm using more images is in progress.