Using the CLEAN algorithm to restore undersampled synthetic aperture sonar images

We present an implementation of the CLEAN algorithm for synthetic aperture sonar. The algorithm is designed to restore image degradation due to gaps in data. We have studied its performance on a variety of scenes based on both simulated and real sonar data. When the scene is composed of point-like targets, the algorithm performs very well, rapidly retrieving the true image that would be derived from fully sampled data. This success can be achieved in spite of a substantial 50% undersampling fraction, corresponding to a platform tow-speed twice as fast as the synthetic aperture limit. However, when the scene is composed of extended bright regions, the algorithm tends to either converge extremely slowly, or fail entirely. Quantitative measures of the rate and degree of restoration are discussed, as well as images before and after the CLEAN process. If the data gaps are caused by towing a sonar platform at a speed in excess of the natural synthetic aperture limit, the successful repair of images derived from undersampled data permits an acceleration in the rate at which the seafloor may be surveyed for mines.