Acoustic Scattering Echo Characteristics And Classification And Recognition Of Underwater Typical Shell Targets

Underwater quiet target detection and recognition is of great significance to offshore defense,port and waterway secutity,which is one of the important research directions in domestic and international field of underwater acoustics.Under the development trend of miniaturization and intelligence of underwater equipment,research on intelligent recognition methods will become the development direction of underwater target detection and recognition.However,the effectively annotated data are limited and difficult to obtain,which restricts the current development of underwater active target recognition.This thesis takes the underwater quiet small target as the research object,aimming at the characteristics of target acoustic scattering echo,the signal feature extraction,classification and recognition in the detection and recognition are researched,and a signal processing scheme for underwater target recognition is proposed,which provides a solution for the intelligent recognition of underwater small targets.Aiming at the spatial inhomogeneity of acoustic scattering field of underwater targets,a simulation model of acoustic scattering echo of underwater targets is established based on acoustic scattering theory.The separation variable method is used to calculate the acoustic scattering field of a typical simple underwater target,and the backscattering morphology function is applied to describe its basic characteristics in the frequency domain.For the hemispherical cylindrical shell that cannot obtain the rigorous theoretical solution of acoustic scattering,COMSOL Multiphysics is applied to establish the target scattering calculation model,and the relationship between the scattering echo charac-teristics and the sound wave incident angle is studied.Then carry out the pool measurement experiment of the hemispherical cylindrical target acoustic scattering,analyze the difference between the target’s acoustic scattering characteristics under the simulation conditions and the experimental measurement results,and provide a theoretical basis for the establishment of the simulation target’s acoustic scattering feature model and target echo feature extraction.Aiming at the problem of coupling interference of acoustic scattering features caused by mixed superposition of target acoustic scattering echo signals in time-frequency domain,this thesis studies the method of extracting high-resolution acoustic scattering features of target echo based on sparse signal representation.Firstly,according to the acoustic scattering mechanism of underwater target,the acoustic scattering echo signal is modeled,and the chirplet dictionary is constructed by combining the frequency modulation of the transmitted signal.Then introduce a sparse signal representation method to reconstruct the echo signal,which takes the time and frequency resolution of multi-component signals into account and extract the time-frequency distribution fatures without cross-term interference.Finally,the extracted time-frequency distribution is transformed by Hough transform and accumulated statistically in the parameter space,and the relationship between the time-series structure characteristics of target echo and the incident angle of sound wave is established.The experimental data processing results show that the research method can obtain high-resolution acoustic scattering characteristics of target echoes,and provide data support for subsequent target classification and recognition.Aiming at the classification problem of underwater elastic spherical shells with different materials and scales,this thesis combines the deep learning method to study the bottom feature extraction and classifier design respectively.The backscattering morphology function is used to describe the acoustic scattering characteristics of targets with different parameters,and the method of parameter scanning and segmented observation is used to establish a target feature database within a certain parameter range,and a deep convolutional neural network suitable for spherical shell classification is constructed.The material classification and geometric parameter estimation of the spherical shell target are realized through the network model.Compared with traditional classifiers and fitting formula methods that use auditory perception features,the deep learning classification model proposed in this thesis has higher classification accuracy and wider application conditions.Aiming at the problem of difficulty in obtaining effectively marked real echo data in underwater target classification and recognition task,a deep learning recognition model based on simulated target echo data training and real target echo data verification is studied.The omnidirectional acoustic scattering echo of a hemispherical cylindrical is simulated,combining the above characteristics to establish different network architectures suitable for underwater target recognition.Train the omni-directional characteristics of the target echo,and use the real target echo data verifies the accuracy of the model and realizes target recognition under different incident angles of sound wave.Further,in the pose classification problem of specific targets,deep neural networks with different structures are designed based on the above-mentioned features to achieve accurate classification of the relative poses of underwater targets.This thesis combines feature extraction technology and deep learning network structure,establishes a signal processing scheme of target classification and recognition,and realizes the mapping from signal features to categories.The research method can provide technical support for the establishment of underwater target recognition system.