Study On Radar Correlative Filter With Generalized Wavelet Neural Network

It is very important to realize target tracking real-timely and accurately in the marine radar signal processing. That represents the level of the signal processing of the system.This paper investigates the prediction problem in the tracking. The real-time processing thinking of nonlinear parallel correlated filter on-line is advanced firstly for marine radar data processing in this paper. It also presents the multi-dimensional processing model with generalized wavelet neural network(GWNN) and its correlated predictive algorithm. The main idea of this paper is to establish the nonlinear parallel neural network model for the processing of radar data by means of the model of feedforward neural network(FNN) and the wavelet theory in signal composition and edge detection. The optimum and adaptive technology is also applied. The main results can be summarized as following:(1) A novel method of selecting the initial optimum step and its adjustment is presented in FNN, which is suitable for the prediction on-line. The misadjustment analysis for the given active function implies the influence of the momentum item.(2) This paper presents the GWNN model and its training algorithm based on the signal composition by S-type wavelet frame. The theory analysis and simulating results show that the convergence and robust property of the GWNN is better with respect to BP net and wavelet net. The active function in the GWNN can be replaced by other frames or spline wavelet.(3) The GWNN with local connection is presented to reduce the scale of the net and to improve the convergence for the multi-dimension GWNN.(4) The nonlinear parallel correlative model with GWNN for marine radar data processing is advanced for the first time. The unique characteristics of the model is that: i)The data processing is not limited by moving pattern of the target. ii)The cost of computation is not associated with the complex of the input. iii)The robust property is obviously better in the prediction of the target with taking a sudden turn or variable acceleration, iv) The GWNN possesses the function of correlative detection real-timely and avoids the delay in the wavelet decomposition.(5) Present the wavelet decomposition algorithm on-line to solve the connecting problem of separate period in wavelet decomposition.