Research On Line Spectrum Detection Technology Based On Hidden Markov Model

Line spectrum feature is one of the main features of underwater acoustic target radiated noise signal.In the actual marine environment,due to the application of vibration and noise reduction technology,the intensity and characteristics of underwater acoustic target radiated noise continue to weaken,and the line spectrum feature is more and more difficult to detect.Focusing on this problem,this paper uses coherent cumulative power spectrum technology and spatial spectrum estimation technology to improve the signal-to-noise ratio in the frequency domain,and uses the hidden Markov model in statistical signal processing to track the target line spectrum first and then detect it in the time-frequency domain,so as to complete the detection of the start and end time of the target line spectrum and whether there is a target or not.The main contents of this paper are as follows:1.In the aspects of Ship Radiated Noise Acquisition and frequency domain processing,the simulation of ship radiated noise is realized,and the background is balanced by polynomial fitting and empirical mode decomposition fitting.Because the traditional periodic graph method and piecewise periodic graph method do not consider the phase information,it will produce actual energy loss and detection error,so the phase compensation method can be used to improve the gain.In this paper,the phase compensation method and its improved method are used to estimate the power spectrum.The comparison shows that this method can improve the processing gain.In order to improve the problem of low frequency resolution,spatial spectrum estimation methods,such as music method,can be used to estimate the frequency.In the frequency domain line spectrum detection,according to the statistical theory,laida criterion and improved laida criterion are used to detect the line spectrum,the improved laida criterion improves the detection performance of line spectrum.2.The traditional time-frequency image acquisition method is short-time Fourier transform,but its frequency resolution is low.This paper uses synchronous extraction transform to improve this problem,and analyzes the advantages and disadvantages of the two methods.In time-frequency domain line spectrum tracking,Viterbi algorithm and forward backward algorithm in hidden Markov model are used for tracking.Under the condition of single line spectrum and multiple line spectra,the two algorithms are compared.The results show that the two algorithms can effectively improve the signal-to-noise ratio.Aiming at the problem of numerical underflow in Viterbi algorithm,Viterbi algorithm is improved and successfully solved.The reliability and effectiveness of the two algorithms are verified by simulation experiments and actual sea trials.3.In the time-frequency diagram,the line spectrum may not exist at all observation times,so it is necessary to detect the existence state of the line spectrum tracked in Chapter 3 within the observation time.In this paper,the sequence test principle is used to detect the start and end time of line spectrum,determine the presence or absence of target in the observation time period,and complete the detection of single and multiple line spectra.Through the processing of simulation and experimental data,the effectiveness of hidden Markov model line spectrum detection is proved.