Study On Identification And Classification Methods Of Whale Acoustic Signals Between Whale Species

Whales can emit different types of sounds in order to accomplish multiple life activities such as underwater target location,individual communication,target recognition,and environmental awareness.The study of whale sounds can assist humans to protect marine mammals and to develop marine living resources.Passive acoustic monitoring of whales is an important technique for whale sounds research.In this process,accurately detecting,identifying and classifying the sounds from different whale species is the premise and basis of the subsequent passive monitoring application,which is of great significance.This paper studies the identification and classification methods of whale acoustic signals between species.The main research contents and completed work are as follows:1.Based on the current situation,development trend and related technologies,the key techniques of whale sound signal recognition and classification methods are analyzed.Based on the time-frequency characteristics of whale sound signals,the overall scheme of the whale click signal and the whale whistle signal identification and classification method are proposed respectively.2.The sound denoising process of the whale sounds identification and classification process was studied.Through the denoising experiment of whale click signal and whale whistle signal,the denoising effects of different methods on the two whale sound signals are studied and analyzed.Sound denoising is accomplished by selecting the appropriate denoising method.3.Based on the traditional endpoint detection method,traditional time-frequency analysis method and traditional classifier,an identification and classification method of whale click signals is proposed.Wavelet transform is applied to the whale click,and the algorithm is designed to adaptively estimate the effective coefficient area in the wavelet coefficient matrix.An adaptive time-frequency feature extraction algorithm is designed to extract the duration and energy distribution characteristics of the whale click signal.Experiments were carried out to verify the effectiveness of the identification and classification methods with sperm whale clicks,long-finned pilot whale clicks,noise interference and artificial sonar signals as experimental data.4.Based on convolutional neural network,an identification and classification method of whale whistle signals is proposed.Based on the powerful data learning ability and feature abstraction ability of convolutional neural network,the whale whistle detection model and the whale whistle classification model are designed respectively,and the adaptive learning of related parameters in endpoint detection,feature extraction and classification is realized.Based on PyQt,the user graphical interface program is designed to realize the visualization of the detection and classification process.The killer whales whistle signals and long finned pilot whale whistle signals were used as experimental data to verify the effectiveness of the proposed identification and classification methods.The experimental results show that the identification and classification methods proposed in this paper obtain more than 95% detection accuracy or classification accuracy rate on the corresponding experimental data sets.The proposed methods can be applied in the fields of passive acoustic monitoring and passive sonar detection.