Surface Target Recognition Based On Unsupervised Learning And Deep Learning Of Vector Sonar

Surface target recognition technology has been the research focus in the field of underwater acoustics.However,due to the complexity of sound field environment and the difficulties in obtaining experimental data and feature extraction,the recognition results of surface targets still fail to achieve the expected results.Therefore,how to recognize the class of target quickly under a high recognition rate is still a problem in surface target recognition technology.A uditory features simulate the auditory principle of human ear,that is,people can distinguish the sound category even in noisy environment.The auditory features can be treated as the features of the surface target.It is necessary to study the neural network based on unsupervised learning to classify the data without labels.In the target recognition methods based on super vised learning,with the extensive application of deep learning in various fields,it can be introduced into the field of underwater acoustics to realize the recognition of surface target without extracting features in advanceThis paper mainly deals with the data received by vector sonar to realize the classification of surface target by combining auditory features with neural network based on unsupervised learning,and the recognition of surface target by neural network based on deep learning.Therefore,this paper first studies the related methods of auditory domain feature extraction were discussed.Combined with the theory of vector signal,the auditory domain feature extraction of vector signal is carried out on the experimental data,and the experimental results are analyzed.Construct three kinds of competitive neural networks based on unsupervised learning and combine the auditory features to classify the surface targets.The three kinds of network models are evaluated by the evaluation criteria of model parameters.The effects of different features,different network models and network parameters on recognition rate are discussed.At last,construct the neural network based on deep learning and the weight sharing in time recursion.Through the processing of experimental data,evaluate the model and analyze the impact of different network models and network parameters on recognition rate.