Analysis Of Weak Target Detection Method In The Chaotic Sea Clutter

A method for detecting weak signal embedded in sea clutter is an important research aspect in the field of signal processing. Target signal is easy to be covered by noise when the detected signal is weak. To extract target signal effectively from the sea clutter has important practical significance.This papers focus on the detection weak signals by additive model and non-additive model.A method for detecting weak signals embedded in chaotic noise by selective support vector machine ensemble based on the theory of phase space reconstruction of the complicated nonlinear system is presented. For improving the generalization ability of support vector machine ensemble, K-means algorithm is used to select the most accurate individual support vector machine from every cluster for ensembling, It is established a One-step predictive model that detects the weak signal, including transient signal and period signals, from the predictive error in the chaotic sequences. It is illustrated in the experiment, which is conducted to detect weak signals from Lorenz chaotic background and IPIX Sea Clutter, that is proposed method is highly effective to detect weak signal from a chaotic background as well as minimize the impact of noise on weak signals, Compare to RBF neural network and SVM models, the new method presents great value in prediction accuracy and detection threshold.With the method of spatial fractal character, target detection can be regarded as a binary-classification, where the clutter-only pattern is available for the classifier design and target detection is to judge whether the received echoes belongs to the clutter-only pattern. For the classification, a feature united detection algorithm based on the non-additive model is proposed in the paper. First, extract two features from the received echoes by multifractal detrended fluctuation analysis, two features are combined to compose two-dimensional characteristic vector. Then, a convex hull training algorithm is utilized to determine a decision region. Finally, the detection rule is whether the decision region surrounds the vector, Experimental results by the raw IPIX radar data show that the proposed algorithm outperforms the compared algorithms.