Detection in Range-Heterogeneous Weibull Clutter Ulku Cilek Doyuran Yalcin Tanik ASELSAN Inc. Middle East Technical University Ankara, Turkey Ankara, Turkey udoyurangZase1sanwcomrtr tanikgmetu.edu.tr Abstract - The paper deals with detection in range-heterogeneous parameter. This model is general and suitable for clutter Weibull clutter. Heterogeneity is defined as a sharp transition interfaces. In this study, assuming that clutter is dominant, the from one homogeneous region to another with a different scale effect of thermal noise is neglected. and/or shape parameter. The proposed method, firstly, tests the The method can be summarized as follows: Firstly, the homogeneity of the window. If it is found to be heterogeneous, CFAR window is identified to be homogeneous or not by the transition point is estimated and the threshold is set accordingly. means of a goodness-of-fit test. Then, all of the samples It is known that usual CFAR methods suffer from target within the window are used in estimation of the shape and masking or increased false alarm rate when operated in non- scaleparametersofthedistribution. Ifthewindowisfoundto Rayleigh or range-heterogeneous clutter. The proposed be heterogeneous, transition point is estimated. In this case, algorithm is shown to perform well in homogeneous or the region of the cell under test is determined by comparing heterogeneous Weibull clutter environment and yield better the position of it with the position of the transition point. performance than commonly known methods. Training samples that are obtained in this way are used in estimating the parameters of the distribution. The parameter I. INTRODUCTION estimates are finally used in determination of the detection In high-resolution radars, since the number of scattering threshold. components is finite or fluctuating in a cell, central limit The paper is organized as follows: In Section II, the theorem is no longer valid and clutter amplitude distribution problem and the clutter model is introduced. The detection deviates from Rayleigh. The most commonly used models to method is described in detail In Section III and performance represent this type of radar clutter are Weibull, log-Weibull, analysis results are presented in Section IV. lognormal and K, which are two-parameter distributions. In this study, Weibull distribution is chosen in modeling of the II. CLUTTER MODEL spatial amplitude distribution of high-resolution radar clutter, The clutter is assumed to have Weibull amplitude as it can fit a wide range of clutter types, over a wide range of distribution and the distribution parameters change abruptly in conditions [1]. Rayleigh distribution is a special form of range. Weibull distribution. The scenario is illustrated in Fig. 1. Here,f(x, a,, /31) is the Common detection schemes that were designed for probability density function of the first clutter region and Rayleigh clutter suffer from increased false alarm rate caused f (x a2, /32) is the one of the second one. Parameters a,, /3, by the spiky nature of this high-resolution clutter. Thus, non- and a2 /32 represent the distribution parameters of region 1 and Rayleigh clutter may be a serious problem. Another difficulty 2, respectively. in detection is that amplitude of radar clutter signal may change in range due to interfaces such as sea-to-land or light- to-dense foliage fields. This situation is problematic for conventional Constant False Alarm Rate (CFAR) detectors f(x a,, /31) f,(x, 2, 182) that were designed for homogeneous clutter environment, since clutter edges may cause target masking at the low- Fig. 1. Clutter PDF transition fromf(x, a,, /3) tof(x, a2, /2) amplitude part of the clutter or false alarms at the transition point. The algorithms designed for homogeneous clutter may The probability density function (PDF) of Weibull yield poor performance even if they were designed for non- pronais d en below: Gaussian clutter. distribution is defined as given below: Many CFAR detectors for Weibull clutter [1],[2],[7] or range-heterogeneous Rayleigh clutter [3],[4],[5],[6] have been f (xla )ex xla suggested before. In this study we design a CFAR method- , a) named Range-Heterogeneous (RH) CFAR- that is suitable for non-Rayleigh and range-heterogeneous clutter. Range whr aan /?aetescladsap prmtr, heterogeneity iS modeled as a sharp transition from one region to another with a different Weibull-PDF scale and/or shape respectively. 1-4244-0284-0/07/$20.OO ©2007 IEEE 343