The Feature Extraction And Classification Of Underwater Acoustic Signal Based On Deep Learning Methods

The complexity of the marine environment,the widespread presence of environmental noise,and the ever-weaker underwater target signal characteristics make underwater target identification a challenging issue.Most traditional approaches for recognizing the underwater targets depend on expert knowledge,so researchers have to extract features from signals manually.However,with the rapid development of sensors and intelligent information technology,these traditional methods have been gradually unable to adapt to the intelligent development requirements of underwater exploration information processing.In recent years,the deep learning method with great success in the commercial field has provided a new technical approach for intelligent underwater target recognition.In this paper,two kinds of typical underwater acoustic signals are taken as underwater targets,which are vessels' radiated noise and whale calls.First,a variety of traditional methods are used to extract the features of these underwater acoustic signals,such as Support Vector Machines(SVM),Error Back Propagation Neural Network(BPNN)and K nearest neighbor algorithm with weights.Second,two kinds of Deep Learning methods are used to construct the classification models,which are unsupervised Deep Belief Network(DBN)and supervised Convolutional Neural Network(CNN).Third,the Transform Learning method is used to train the CNN,and the similarity of the whale calls' samples is analyzed based on the abstract features extracted from the CNN.At the same time,the phylogenetic tree of whales is drawn and the ability of the “abstract features” is analyzed based on it.The results show that the Deep Learning method adopted in this paper can effectively extract the "abstract features" of underwater acoustic signals.Compared with the traditional classification and recognition methods,the models of classification and recognition have obviously improved both in recognition accuracy and intelligence.