Feature Extraction And Classification Of Cetacean Acoustic Signals Based On Sparse Representation

The ocean covers 70.8%of the earth's surface area and is the largest ecosystem of the earth.Cetaceans are the most common and representative group of marine mammals.The species and quantity of cetaceans are important biological indicators for monitoring the quality of marine environment.Cetacean calls have the characteristics of species,which are the important biological characteristics of cetaceans,also the important basis of classification and recognition.The research object of this thesis is underwater acoustic signals of cetaceans.The research goal is to extract more accurate feature of useful signals in cetacean acoustic signals,design a reasonable classification and recognition scheme,and further improve the accuracy of classification and recognition of cetacean acoustic signals.Based on the analysis of the existing whale classification and recognition technology,this thesis studies the feature extraction and classification of whale acoustic signals based on sparse representation theory,which provides technical support for whale species monitoring and marine ecological environment monitoring,and has theoretical reference significance for the research of bionic technology in underwater acoustic communication.This study has certain economic benefits for the national marine industry,and has a broad application prospect in the intelligent marine ecological industry.The main contents and innovations are as follows:1.Introduce the theoretical knowledge of whale acoustic signal classification and recognition.This thesis discusses the characteristics of cetacean acoustic signals,and focuses on the main feature extraction algorithms and the main advantages and disadvantages of classifiers in this field.2.Based on the resonance sparse signal decomposition(RSSD)and the combination of Mel frequency cepstral coefficients(MFCC),syllable length and four entropy,we proposed a new whale sound signal classification and recognition method.By making full use of the oscillation properties of various components in whale sound signals and selecting the quality factors of different rotatable Q-factor wavelet transform,the whistle signal components with high degree of oscillation and the signals with low degree of oscillation(such as environmental noise)are effectively separated,so as to directly extract the whistle signal components.In addition,after extracting whistle signal components,MFCC features is extracted in the frequency domain,four entropy features and the sound signal length of each sound are extracted in the time domain as the recognition basis of support vector machine(SVM)and random forest(RF)classifiers.Experimental results show that our method is better than other methods,and further improves the recognition accuracy.This method provides an effective analysis method for whale acoustic signal classification and recognition.3.Based on the tunable Q-factor wavelet transform(TQWT)and the basis pursuit de-noising(BPD),we proposed a new method for extraction of whistles and clicks from cetaceans.According to the different oscillation properties of whistle component and click component in cetacean sound signals,we use TQWT algorithm to extract the whistle component and click component respectively,adopt basis pursuit de-noising method to remove the noise.Our new method provides an effective analysis method for Cetacean signal classification and recognition.4.Based on RSSD and ridge extraction,we proposed a new method of whale acoustic signal classification and recognition.Optimize the different oscillation properties of various components in cetacean acoustic signals,we extract the components of whistle signals from the original signals by RSSD algorithm.Then,we extract the ridge of whistle in time-frequency domain.At last,we extract the ridge features and employ support vector machine and random forest to classify.The classification results show that the performance of this method is particularly excellent,and the classification and recognition rate of all four cetaceans is over 98%.