Source Localization Method Research In A Unceutain Waveguide Environment

The ocean is a bounded waveguide environment,and the traditional method for sound source localization in the ocean is Matched field processing(MFP),which uses an acoustic propagation model and array reception data for matching to localize the sound source,and can greatly improve localization performance compared to plane-wave beamforming-from direction finding to distance-depth estimation.-from direction finding to distance-depth estimation.However,the performance of conventional MFP relies heavily on accurate waveguide environmental parameters and acoustic propagation modeling techniques,but in practice,the relevant parameters are difficult to measure accurately,and the spatial and temporal variations in the ocean interior make the waveguide parameters uncertain,so the practical application of MFP is very difficult.In this thesis,three tolerant underwater sound source localization methods are proposed for the case of waveguide parameter uncertainty,and their localization capability in the uncertain waveguide environment is investigated.The research in this thesis is centered on sound source localization in the uncertain waveguide environment,and the central idea is to gradually increase the importance of the data,while mining the source location and waveguide parameter information from the data,and then realize the tolerant underwater sound source localization.The research is as follows:First,a Bayesian matching field processing method is proposed for the environmental parameter uncertainty problem,which models the uncertainty of environmental parameters based on array data and Bayesian principle,and then does posterior probability weighted summation with linear and adaptive matching field processors as primitives,and simulation studies show that the method can significantly improve the tolerance of matching field processors.Secondly,to address the problems of high computational effort and high parametric level of Bayesian matching field,an adaptive matching field localization method based on covariance matrix fitting is proposed to transform the MFP localization problem into a problem of maximizing the output power under the constraint of the range of copy vector uncertainty by introducing the set of nonrepresentative copy vectors,using Lagrange multiplier method to find the copy with the highest degree of fit to the array covariance matrix The Lagrange multiplier method is used to find the copy vector with the best fit to the array covariance matrix to estimate the spatial power spectrum of the source.Simulation studies show that the method can obtain reliable localization performance in the uncertain waveguide environment by selecting appropriate regularization parameters,and the computational effort and parametric level are significantly reduced compared to the Bayesian matching field processing method.Finally,in order to further enhance the importance of data,and considering that the acquisition and processing of hydroacoustic labeled data is more restricted than other scenarios,while unlabeled data is relatively easy to obtain,a semi-supervised learning classification model-based underwater sound source localization framework is proposed,which is a "model-free" data-driven approach.The main body consists of two parts: a residual attention mechanism convolutional self-encoder trained in unsupervised mode and a multilayer perceptron trained in supervised mode.The method is validated on a uniform vertical array of SWell Ex-96 event S5 data,and the results show that the proposed underwater acoustic source localization framework can maintain better localization performance than MFP with a smaller set of labeled data,and is tolerant to environmental mismatch(which can also be interpreted as the generalization ability of the model).