Feature Extraction Method Of Target Radiated Noise Based On Space-Time-Frequency Joint Processing

Line spectrum feature is one of the main characteristics of underwater acoustic target radiation noise signal.The detection and extraction of line spectrum features is an important part of underwater acoustic signal processing.The application of the shock absorption and noise reduction technology makes the radiation noise intensity and characteristics of the underwater acoustic target continue to weaken.In the actual marine environment,the line spectrum features are more and more difficult to be detected.At the same time,however,the detection of weak targets is more and more dependent on the detection of line spectrum features.In the multi-beam data acquired by the array,the detection and tracking of the line spectrum needs to be carried out in time,frequency and spatial dimensions,extending the conventional The line spectrum detection and tracking processing dimensions greatly increase the amount of calculation,which brings difficulties to real-time processing.In this paper,line spectrum detection based on power spectrum,LOFAR spectrum and multi-beam LOFAR spectrum data is studied for target signals acquired by sensors and arrays.Firstly,this paper discusses the background and current situation of the research,introduces the generation and mechanism of ship radiated noise,and combs the basic methods of acquiring underwater acoustic signals and extracting spectral features.Secondly,six kinds of fitting methods are used to analyze the background spectrum of the signal spectrum,and the advantages and disadvantages of the fitting method are judged by setting two characteristic parameters of goodness of fit and line energy preservation.Based on the background equalization of the radiation noise spectrum of the ship,the improved local 3? criterion is used to detect the single-frame line spectrum,which improves the performance of the single-frame line spectrum detection algorithm.Thirdly,in the LOFAR graph line spectrum detection research,the LOFAR graph line spectrum detection method based on the birth and death process is proposed,and the line spectrum detection and tracking in the LOFAR graph is realized.Aiming at the low-signal-to-noise ratio line spectrum signal,a HMM-based LOFAR line spectrum detection method is proposed,and the calculation formula of the observation probability matrix elements in the three elements of HMM under the condition of unknown signal-to-noise ratio is derived.The high false alarm detection is used to preprocess the data set to reduce the false alarm probability of the HMM model detection.The method of gridding processing is proposed,which greatly reduces the computational complexity of the algorithm compared with the conventional HMM model line spectrum detection method.Finally,aiming at the problem of line spectrum detection for multi-beam LOFAR data,the method of line spectrum detection based on birth-death process and HMM is extended,and the spatial dimension processing is added,including :(1)Two methods,interference band zero method and difference method,are proposed to eliminate the false alarm of main and side lobes,and eliminate the spatial signal interference caused by the directivity function of array.(2)A dimension reduction calculation method is proposed,which can reduce the dimension of three-dimensional data to two-dimensional data,and make use of K-means algorithm to judge outliers in beam domain and frequency domain to ensure the correctness of the results.(3)A beam pre-screening method is proposed.By counting the number of pre-screened frequency points in each beam,the non-spectral beams are eliminated,which greatly reduces the computational complexity and improves the efficiency of line spectrum detection.