Multiple Change Point Detection-based Target Detection in Clutter

Abstract

The effectiveness of target detection methods in radar systems depend on how accurately clutter can be characterized. However, depending on application, clutter statistics vary, and therefore it is difficult to accurately predict such statistics and their parameters. Model-based detection algorithms that are developed for one clutter scenario will fail to yield satisfactory results in another scenario. In this paper, we propose a complete data driven multiple change point detection (CPD) for target detection which does not requires the knowledge of the underlying clutter distribution. The key concept is to iteratively search for slow time instance that maximizes the cumulative sum (CUMSUM) Kolmogorov-Smirnov (KS) statistics. If such statistics exceeds a pre-specified threshold value, then this slow time instance is added to the collection of the estimated change points. This process continues until all CUMSUM-KS statistics are below the threshold. Computer simulations are used to demonstrate the effectiveness of this method for different clutter distributions.