全光纤安防系统毕设设计.docVIP

摘要ABSTRACT All-optical security system is a distributed fiber-optic disturbance location and sensor defense system, able to obtain the measured changes in space and time in the region. The paper dose research on pattern recognition of fiber security-monitoring system . Firstly, wavelet denoising method and Spectral subtraction method are introduced.?Applicable environment for two methods are given through experiments; Second, the energy characteristic method based on wavelet packet decomposition(WPD), ?marginal spectrum characteristic method based on Hilbert - Huang transform (of HHT) ,method based on bispectrum estimation and a new method to improve signal eigenvectors are proposed.?Then the paper introduces neural networks and support vector machine;?At last this paper presents the whole pattern recognition in combination with actual application. In feature extraction part, this paper presents a method to extract 32-D feature vector using WPD,a method to construct a 51-D?marginal spectral characteristics by HHT, a method to get a 128-D spectrum characteristics using bispectrum estimation.? In the design part of the classifier part, this paper introduces neural network theory and support vector machine theory, proposes a new method combining neural network and support vector machine strategy. The results show that the wavelet energy feature method, three improved marginal spectrum characteristic method and the bispectrum characteristics method can recognize three different fiber disturbance signals accurately.?The wavelet method and bispectrum method can be real-time, accurate, highly-effective. For bispectrum estimation method spends more time and presents better accuracy. Compare with support vector machine method, neural network classifier is better for combining bispectrum feature extraction.The recognition rate is 100% for the test samples.? Key words: pattern recognition; feature extraction; wavelet package decomposition; hilbert-huang translation; ?bispec