Dissertation
Dissertation > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Communication theory > Signal processing

Research on Underwater Target Recognition

Author YuQiuXing
Tutor LiZhiShun
School Northwestern Polytechnical University
Course Signal and Information Processing
Keywords chaos fractal wavelet analysis nonlinear acoustic signal processing feature extraction target recognition classification state space reconstruction principal component analysis (PCA) fractal dimension generalized fractal dimension wavelet neural network.
CLC TN911.7
Type PhD thesis
Year 2004
Downloads 1019
Quotes 12
Download Dissertation

Recognition technique on underwater targets has important value not only in economic field, but also in military field. Most countries in the world pay much attention to it. Although more and more novel methods are proposed and employed in this field, such as chaos theory, fractal theory, and wavelet theory, most of them are nearly on the beginning stage. The key of the recognition technique can hardly be obtained from abroad for some reasons. This dissertation focuses on extracting features from underwater acoustic signals and classifying them based on chaos & fractal signal processing and wavelet theory. Following are the primary contributions.1) The theories of chaos & fractal, wavelet and involved concepts of target recognition are discussed in detail. The mathematical definition of chaos and the two parameters that can describe chaotic behavior are introduced. And wavelet analysis is widely used for its good localization both in time domain and in frequency domain, thus the concepts and principles of discrete wavelet transform, multi-resolution analysis (MRA) and wavelet packet analysis (WP) are presented. Finally, the process of target identification is depicted, which includes five parts.2) Chaos theory is systemically expounded and its forming and development are summarized. Meanwhile, the definitions and computation of strange attractor’s fractal dimension and the Lyapunov exponent are discussed when the chaotic property is quantitatively analyzed.3) State space reconstruction, which is an essential method in chaos signal processing, is analyzed in detail, and involved parameters are seriously selected. Then the principle and algorithm about reducing noise by principal component analysis (PCA) are discussed, and the simulations in different signal-noise-radio (SNR) show that PCA can effectively reduce add-in noise. At the same time, some features are extracted from five kinds of real echoes by PCA.4) A novel method based on state space reconstruction and K-L transform is proposed, which improves the method in reference 12. The simulation results show that the extracted features are robust, and the recognition accuracy is satisfied.5) Applying wavelet theory to feature extraction in wavelet domain is studied. The methods based on MRA and WP are discussed, and the ways to form the feature vectors are proposed. The influence of scales, mother wavelet, anddimensions of features is discussed.6) A feature extraction approach based on wavelet and fractal theory is developed. Compared the feature extracted by MRA with that by the approach, the simulation results that average correct recognition rate rises from 91.67% to 95.34% show that recognition accurate is improved.7) The combination of MRA and fractal theory is studied. Inspired by the fractal idea, wavelet multi-resolution fractal features, which belong to the generalized fractal dimension feature, are obtained. Extracting the features from five kinds of real echoes and classifying them are accomplished, and the results demonstrate that the features are feasible. Therefore, a new method and new features are acquired, which are helpful to recognize the echoes.8) The concept, principle and algorithm of wavelet neural network are explained. And the method to select the initial parameters is developed. Meanwhile, the wavelet neural network is employed to compress and classify the underwater echoes.

Related Dissertations
More Dissertations