Detection And Extraction Of Weak Target In Sea Clutter Based On RBF

With the increasing development of radar technology，the study on sea clutteris becoming more and more thorough.Target detection and extraction of seaclutter will become an very important role in sea surface radar signal processing,which can provide basis to small and low target flying on the sea in the militaryand provide accurate information on the location and the target for the missileforces, provides accurate position information for offshore fishing, sea rescue andsmuggling in the civil area and provide security for our sea barrier. The chaoticcharacteristic of sea clutter is the main line of research.The RBF neural networkone step prediction error detection method has been used in sea clutter targetdetection mainly utilizing the RBF neural network’s good prediction on chaotictime series. The choice of CFAR detector is vital to the target detection andextraction of sea clutter. According to the CFAR theory, the CFAR detectorclassification requirements and CFAR hold, research and analysis of the chaoticcharacteristics of sea clutter and verifying CFAR detector improvement predictionerror method detection model efficacy will be the main efforts in this paper.Many researchers over the world have inferred that the sea clutter has chaoticcharacteristics. The chaotic characteristics of sea clutter will be studied further andsmall amount of data method is used to estimate the the largest Lyapunovexponent. And for the first time, we use power-weighted method to calculate seaclutter time series average period and linear region obtained by Max no fluctuationinterval method. The proposed method can improve the speed and accuracy of thesea clutter largest Lyapunov index calculation. The RBF neural network stepprediction model is built to predict sea clutter time series. Through the analysis ofthe structural parameters of RBF neural network, RBF neural network hiddenlayer expansion coefficient is adjusted, the sea clutter time series predictionperformance is improved. Target detection and extraction of sea clutter isaccomplished by using RBF neural network step prediction error data based on seaclutter time series with chaotic character. RBF neural network one step prediction error detection model is established. We use the improved CFAR detector toanalyze chaotic characteristics of sea clutter and conduct target detection applyingRBF neural network step prediction error data. The CFAR detector is theimprovement of the orderly class CFAR detector principle. The proposed CFARdetector don’t affect the CFAR loss with the change of the amplitude and at thesame time, it can maintain a good false alarm rate in detecting weak targets. Themain works in this paper are organized as follows:Firstly,The maximum Lyapunov index, correlation dimension andKolmogorov entropy of sea clutter are used to determine whether the sea clutterhas the chaotic character. By using the real data IPIX#17of sea clutter posted onthe McMaster page, the maximum Lyapunov index of different sea clutter dataare calculated and analyzed by using Wolf method and small-data method.Correlation dimension is calculated by using G-P algorithm, the Kolmogoroventropy is obtained by using maximum likelihood estimation method. Make surethe obtained sea clutter data have chaotic character.According to the structure andtheory of RBF neural network, the hidden layer parameters(expansion coefficient)are adjusted and analyzed to reduce the training error, promote the performance ofsea clutter time series prediction.Secondly, according to the view that sea clutter has chaotic characteristics,hebasic theory and characteristics of chaos are analyzed. Since the RBF neuralnetwork for chaotic time series with a good fitting characteristics, RBF neuralnetwork one step prediction error detection model is established. The time seriesof sea clutter is predicted and model is verified using Lorenz time series. Chaotictime series phase space reconstruction use C-C method, mutual Informationmethod and cao method. After the analysis comparisons of delay time andembedding dimension, the accurate will be obtained.Thirdly, The basic functions of average unit CFAR detector and the orderedclass CFAR detector are analyzed and find out there is insufficient in bothmethods. The CFAR detector improved order statistics CFAR detector is used toanalyze the function of the proposed detector. Through the structural analysis andthe principles performance indicators demonstrate that improved detector meetCFAR detector requirements.The CFAR detector of RBF neural network one stepprediction error model is established. The probability density function of the CFAR forecast detection unit error data is calculated by using Kernel functionestimation method. The threshold of the target is obtained by probability densityfunction and the corresponding CFAR. The rectangular target signal containing achaotic background and single pulse target signal containing a chaotic backgroundare used to validate the proposed model. Finally, the proposed model is used toperform the target detection and extraction of sea clutter.