Integration of Fine Model-Based Decomposition and Guard Filter for Ship Detection in PolSAR Images

Dongsheng Liu,Ling Han

doi:10.3390/s21134295

Abstract

Ship detection with polarimetric synthetic aperture radar (PolSAR) has gained extensive attention due to its widespread application in maritime surveillance. Nevertheless, designing identifiable features to realize accurate ship detection is still challenging. For this purpose, a fine eight-component model-based decomposition scheme is first presented by incorporating four advanced physical scattering models, thus accurately describing the dominant and local structure scattering of ships. Through analyzing the exclusive scattering mechanisms of ships, a discriminative ship detection feature is then constructed from the derived contributions of eight kinds of scattering components. Combined with a spatial information-based guard filter, the efficacy of the feature is further amplified and thus a ship detector is proposed which fulfills the final ship detection. Several qualitative and quantitative experiments are conducted on real PolSAR data and the results demonstrate that the proposed method reaches the highest figure-of-merit (FoM) factor of 0.96, which outperforms the comparative methods in ship detection.

Highlights

IntroductionSince polarimetric synthetic aperture radar (PolSAR) can offer intuitive physical explanations for scattering behaviors that could make targets or structures identifiable, ship detection with PolSAR images has received continuous attention and is crucial for maritime surveillance [1,2,3]
Importantly, detection of ships with small dimensions and weak backscattering in rough sea state scenarios is always troublesome. To deal with these issues, this paper proposes a polarimetric synthetic aperture radar (PolSAR) image ship detection method through integrating the fine model-based decomposition (MBD) and guard filter
This work contains the following aspects; first, to accurately describe the local structure scattering of ships, four advanced scattering models are introduced and an eight-component decomposition is put forward; second, by analyzing the scattering mechanism differences between sea clutter and ships, a novel detection feature is designed based on the derived scattering contributions; third, considering the utilization of spatial information, a guard filter is constructed and incorporated to further amplify the efficacy of the feature; experiments are conducted on real PolSAR data to qualitatively and quantitatively assess the proposed method with comparative methods and the results demonstrate that the proposed method enjoys optimal detection performance and can effectively enhance the targetto-clutter ratio (TCR)

Summary

Introduction

Since polarimetric synthetic aperture radar (PolSAR) can offer intuitive physical explanations for scattering behaviors that could make targets or structures identifiable, ship detection with PolSAR images has received continuous attention and is crucial for maritime surveillance [1,2,3]. Based on the statistical analysis, the constant false alarm rate (CFAR) is one of the popularly used techniques for ship detection [4,5]. In the case of unknown prior information, the CFAR enjoys favorable detection performance since ships have more intense scattering responses compared with sea clutters. An emerging method is to train a deep neural network to distinguish the important features. In this vein, the convolutional neural network (CNN)

Results

Discussion

Conclusion