Weak Targets Detection In Sea Clutter

Radar detection techniques for small target in sea clutter have been used in port traffic, ocean monitoring, military reconnaissance, maritime and air rescue, etc.. As the radar signal of small target in sea clutter is very weak, and is nonstationary, it is a difficult task to detect weak target. The core of small target radar detection is how to accurately and effectively extract radar target information in the strong sea clutter background. Generally, the weak object detection needs to choose the appropriate feature vectors and the classifier.This thesis selects a combined feature which reflects the sea clutter characteristics more precisely. The sea clutter databases are first preprocessed by three parameter fractional Fourier transform. Then, time and frequency domains characteristics of sea clutter data, including Hurst exponent, Lyapunov exponent, fractal dimension, multi-fractal spectrum, correlation time and approximate entropy, are studied, and a new joint feature vector is obtained by Genetic Algorithm (GA). Finally, a kind of classifier using the Deep Belief Network (DBN) is designed, where the DBN is used to train the features, and map it into a feature space. After the network is trained to achieve the desired effect, it is used to detect and classify the weak object in the sea clutter.The main works of this thesis are as follows:(1) The TFRFT is proposed, which makes the characteristic difference between the sea clutter sub target unit and the main target unit in TFRFT domain further expand.(2) A new joint feature vector is obtained by GA, which reflects the sea clutter information more precisely.(3) The DBN-HMM is used for the classification of sea clutter signal, and the detection accuracy is improved effectively.In order to verify the performance of this algorithm, IPIX database are used for test. The detection accuracy reach 96.33%, and is higher than that of existing algorithms under low SNR. Obviously, the method can improve target detection accuracy.