Enhancing Coherence Images for Coherent Change Detection: An Example on Vehicle Tracks in Airborne SAR Images

Horst Hammer,Silvia Kuny,Antje Thiele

doi:10.3390/rs13245010

Abstract

In Synthetic Aperture Radar (SAR) interferometry, one of the most widely used measures for the quality of the interferometric phase is coherence. However, in favorable conditions coherence can also be used to detect subtle changes on the ground, which are not visible in the amplitude images. For such applications, i.e., coherent change detection, it is important to have a good contrast between the unchanged (high-coherence) parts of the scene and the changed (low-coherence) parts. In this paper, an algorithm is introduced that aims at enhancing this contrast. The enhancement is achieved by a combination of careful filtering of the amplitude images and the interferometric phase image. The algorithm is applied to an airborne interferometric SAR image pair recorded by the SmartRadar experimental sensor of Hensoldt Sensors GmbH. The data were recorded during a measurement campaign over the Bann B installations of POLYGONE Range in southern Rhineland-Palatinate (Germany), with a time gap of approximately four hours between the overflights. In-between the overflights, several vehicles were moved on the site and the goal of this work is to enhance the coherence image such that the tracks of these vehicles can be detected as completely as possible in an automated way. Several coherence estimation schemes found in the literature are explored for the enhancement, as well as several commonly used speckle filters. The results of these filtering steps are evaluated visually and quantitatively, showing that the mean gray-level difference between the low-coherence tracks and their high-coherence surroundings could be enhanced by at least 28%. Line extraction is then applied to the best enhancement. The results show that the tracks can be detected much more completely using the coherence contrast enhancement scheme proposed in this paper.

Highlights

One of the most striking differences of Synthetic Aperture Radar (SAR) sensors when compared to optical sensors is the coherent image formation process, and the ability to measure both amplitude and phase of the return signal
The success of the algorithm is demonstrated by extracting the tracks both from the original and from the optimized coherence image, showing that the latter are much more complete than the tracks extracted from the original coherence image
This algorithm consists of amplitude speckle filtering, removal of the topographic phase and phase noise filtering, which is only performed in the high-coherence areas

Summary

Introduction

One of the most striking differences of Synthetic Aperture Radar (SAR) sensors when compared to optical sensors is the coherent image formation process, and the ability to measure both amplitude and phase of the return signal. The recorded phase, if used in an interferometric image pair, carries a wealth of information about the scene. This information has been exploited in many ways, be it Persistent Scatterers [1], the creation of worldwide digital elevation models [2] or small baseline interferometry [3]. All of these approaches to exploit the interferometric phase rely on the data to be stable, i.e., coherent. Applications of this kind of change detection (Coherent Change Detection, CCD) are manifold and cover

Objectives

Results

Discussion

Conclusion