Research On Feature Extraction And Recognition Technology Of Underwater Acoustic Target Radiated Noise

As a part of underwater target recognition,the underwater acoustic target radiation noise feature extraction and recognition technology is a prerequisite for concealed attacks,first enemy discovery,and the right to fight for war in future submarine warfare and antisubmarine warfare.It is also the realization of other underwater weapon system intelligence.One of the key technologies.In recent years,with the development and application of ship vibration and noise reduction technology,the target radiated noise power is greatly reduced,resulting in many factors such as unclear,unstable and characteristic variation,multi-target interference,and environmental noise interference.The obtained target radiated noise is no longer pure,and the power spectrum of the signal is also doped with many "pseudo features" that are not the target itself,which brings great difficulty to the detection and recognition of passive sonar targets.At present,passive sonar target recognition has become a bottleneck problem affecting anti-submarine warfare.How to extract target features from underwater acoustic target radiated noise,identify interference features,improve the accuracy and reliability of extracted features,and obtain effective features of target radiated noise.It is a research topic that has important significance and needs to be solved urgently.Aiming at the target recognition problem of passive sonar system,this paper studies the feature extraction and target recognition technology based on underwater acoustic target radiation noise.The main work and innovations of the thesis are as follows: 1.The high-fidelity line spectrum extraction technique of target radiated noise under the condition of towed complex line array distortion is studied.A new method of distortion composite line array self-correction based on strong line spectrum delay estimation is proposed.This method only requires target radiation.The strong line spectrum information in the noise enables distortion correction of the composite line array to obtain high fidelity line spectrum characteristics without additional reference sources.Aiming at the target feature extraction problem of multi-objective and strong interference problems in real underwater acoustic environment,a new method for extracting line spectrum features based on multi-objective and strong interference in distorted composite array is proposed.The method is combined with distortion composite array.The correction technique and the generalized sidelobe canceller technology realize high-fidelity extraction of the spectral characteristics of the target signal under strong interference.Finally,the effectiveness of the above method is verified by simulation experiments and sea trial data.2.The multi-band underwater acoustic target modulation spectrum reconstruction technique is studied.A new method of high-resolution underwater acoustic target modulation spectrum reconstruction based on sub-band sparse structure is proposed.This paper introduces the shortcomings of the traditional Fourier transform-based modulation spectrum feature extraction method,and analyzes and utilizes the group appearance structural characteristics and sparse distribution characteristics of the underwater acoustic target radiation noise modulation spectrum features.The modulation spectrum problem of the transform is transformed into the inverse Fourier basis based coefficient reconstruction problem.A nonparametric sparse Bayesian framework for high-resolution modulation spectrum reconstruction is proposed.The Monte Carlo sampling is used.The technology realizes the derivation of the model and obtains the posterior distribution of the probability model.The performance of the algorithm is verified by the simulation data and the sea trial data.3.The existing classification and recognition techniques of underwater acoustic targets and deep learning theory are studied.A new method of spectrum identification based on deep learning for radiated noise signals is proposed.The convolutional neural network is used to classify the time-frequency joint domain features of different underwater acoustic target signals,and the corresponding convolutional neural network is designed and constructed.The output of each layer in the network is analyzed,and the convolutional neural network is used to extract the underwater sound.The effectiveness of the deep-level features of the target radiated noise is verified by the simulation data and the measured data respectively.