Small Target Detection Based On The Chaotic Characteristic Of Sea Clutter

Small target detection in sea clutter plays an important role both in the military andcivilian. Sea clutter is the result of radar backscatter from sea surface which interferes inperformance of sea targets detection seriously. For a long time, sea clutter is modeled as arandom process. With further research, it is discovered that sea clutter has chaotic dynamiccharacteristics, which provides a new method for the detection of small target in sea clutter.The main contents of this paper are as follows:Firstly, the chaotic dynamic characteristics of sea clutter are analyzed in the paper. Phasespace reconstruction is the first step to study the chaotic characterization of sea clutter, whichplays an important role in the calculation of chaotic invariants of sea clutter. The phase spaceof the sea clutter data is reconstructed according to Takens’ embedding theorem. There aretwo important parameters for the phase space reconstruction: the time delay and theembedding dimension. The method of autocorrelation is used to estimate the time delay andGP method is used to estimate the embedding dimension. Correlation dimension and thelargest Lyapunov exponent, which are two important values of chaotic system, are used forthe chaotic characteristics identification. GP method is used to calculate the correlationdimension and Wolf method is used to estimate the largest Lyapunov exponent.Secondly, a new method for small target detection is proposed when the chaoticcharacteristic of sea clutter is verified. This algorithm is based on chaos theory and phasespace reconstruction theory. The RBF neural network is used to reconstruct the nonlineardynamic model of sea clutter and predict the sea clutter time series. In the last step, a newmethod based on prediction error is introduced to detect small target.Finally, genetic algorithm is introduced to optimize the RBF neural network. Accordingto the global search capability of genetic algorithm, a new method is presented for sea cluttertime series prediction using RBF neural network based on genetic algorithms. The parametersof the RBF neural network contain the centers and widths of the radial basis function and theweights between the hidden layer and the output layer. The prediction performance isimproved based on the optimized RBF neural network.