Research On Feature Extraction And Recognition Techniques Of Underwater Acoustic Target

Underwater acoustic target recognition is one of the key technologies in modern naval warfare.It is an important symbol of the intelligence of underwater acoustic equipment and weapon system,and has important military application value.However,the complexity and variability of the marine environment have a serious impact on the stability of the underwater acoustic system,which leads directly to the performance degradation of the identification system.In the field of underwater acoustic target recognition,feature extraction and target recognition techniques are two kinds of core technologies.Therefore,the selection of robust feature extraction and recognition methods is one of the most important research contents in the field of underwater acoustic target recognition and has important practical significance.In this paper,the basic principle of underwater acoustic target recognition is introduced in detail,and the feature extraction method of the underwater acoustic target and the design and selection method of the classifier are mainly studied.In the aspect of feature extraction,this paper mainly studies the feature extraction method of underwater acoustic target based on the auditory feature and the feature extraction method based on the Hilbert's marginal spectrum.Inspired by the auditory perception mechanism of sonar solider,this paper firstly studies the feature extraction method based on Mel frequency cepstral coefficient.Then,aiming at the deficiency of the method,this paper further proposes the feature extraction method based on Gammatone frequency cepstral coefficient and the feature extraction method based on Gammatone filterbank and subband Instantaneous Frequency.Because of the unique advantages of Hilbert-Huang transform in the analysis and processing of non-stationary signals,this paper applies the Hilbert-Huang transform to the feature extraction of underwater acoustic signals,and proposes the feature extraction method based on the Hilbert's marginal spectrum.The proposed methods are used to extract the features of the measured ship radiated noise data,and the experimental results show the effectiveness of the proposed feature extraction methods.In the aspect of recognition,this paper mainly studies two kinds of classifiers,BP neural network and support vector machine.This paper introduces their basic principles and algorithm implementation in detail,gives the corresponding parameter selection method,and the advantages and disadvantages and the comparative conditions of the two classifiers are given.Finally,the BP neural network and the support vector machine(SVM)are used to identify the radar noise data of the measured ship.The experimental results further validate the validity of the feature extraction algorithm and the classifier model.At the same time,according to the recognition accuracies under different SNR conditions,the performance of various feature extraction methods are compared and analyzed.The recognition results show that the feature extraction and recognition methods proposed in this paper can effectively improve the performance of underwater acoustic target recognition system and have strong robustness.