Signal Processing 87 (2007) 2100–2110 A novel threshold optimization of ML-CFAR detector in Weibull clutter using fuzzy-neural networks Amar Mezache, Faouzi Soltani à De´partement d’Electronique, Faculte´des Sciences de l’Inge´nieur, Universite´de Constantine,Route d’Ain El Bey, Constantine25000, Algeria Received 1 September 2006; received in revised form 18 February 2007; accepted 19 February 2007 Available online 27 February 2007 Abstract This paper provides a novel and effective approach based on an adaptive neuro-fuzzy inference system for the solution of constant false alarm rate (CFAR) detection for Weibull clutter statistics. The optimal detection thresholds of the maximum-likelihood CFAR (ML-CFAR) and the Censored ML-CFAR (CML-CFAR) detectors in Weibull clutter with unknown shape parameter are obtained using fuzzy-neural networks (FNN) technique. The theory of the FNN is presented and the genetic learning algorithm (GA) is applied for the training of the FNN threshold estimator. The proposed FNN-ML-CFAR and FNN-CML-CFAR detectors proved to be efficient particularly in the case of spiky clutter. Experimental results showed the effectiveness of an adaptive neuro-fuzzy threshold estimator under different system conditions and it is also shown that the optimal FNN-ML-CFAR and FNN-CML-CFAR detectors can achieve better performances than the conventional ML-CFAR and CML-CFAR algorithms. r 2007 Elsevier B.V. All rights reserved. Keywords: Fuzzy-neural networks; Genetic algorithms; CFAR; Weibull clutter 1. Introduction Constant false alarm rate (CFAR) detectors are used to regulate the probability of false alarm ðP FA Þ to a desired level in varying background environ- ments. The Weibull probability density function (pdf) is known to represent sea and ground clutter at a grazing angle of 3.91 or at high-resolution situations [1,2]. When the grazing angle increases, the shape parameter of the Weibull pdf increases and the distribution becomes almost Rayleigh at high grazing angle of 61.41. Some works in CFAR detection for Weibull clutter with unknown shape parameter have been reported in the literature [3–5]. In [3], Ravid et al. developed the maximum-likelihood CFAR (ML-CFAR) algorithm and analysed its performance for the case where both the scale and shape parameters are unknown. Lops et al. [4] introduced a logarithmic transformation to reduce the Weibull pdf to a Gumbel pdf to obtain equivariant parameter estimation and they analysed a CFAR algorithm for unknown scale and location parameters. In the ML sense, the optimal Weibull CFAR (OW-CFAR) is proposed in [5] when the test statistics is expressed according to the estimator of the mean power of the Weibull clutter. They also studied CFAR detection in the case of unknown scale and shape parameters. ARTICLE IN PRESS www.elsevier.com/locate/sigpro 0165-1684/$ - see front matter r 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.sigpro.2007.02.007 à Corresponding author. Tel.: +213 31 90 95 68. E-mail address: f.soltani@caramail.com (F. Soltani).