Cetacean Call Signal Detection And Identification Technology In Polar Ice Areas

The Arctic is one of the most sensitive regions to global climate change,and with human research into the exploitation of resources in the Arctic,anthropogenic emission acoustic signals,environmental noise signals,and ship signals are all acoustic signals that are closely associated with the polar environment and resources,in addition to which vocal signals from large marine mammals such as whales also serve as strong environmental indicators.The study of cetacean acoustic signals contributes to the understanding of the polar acoustic environment.Passive acoustic monitoring is an important technical tool for the study of cetacean signals.However,the Arctic is a complex environment with a wide variety of background noise.Cetacean signals are hidden in vast amounts of data and traditional energy detection methods cannot be identified automatically in situ,as well as manual detection methods are inefficient and unfeasible.Therefore,automated detection and classification methods are needed to improve efficiency and accuracy.Firstly,to address the above issues,this paper proposes an automated detection algorithm based on differential local conditional peaks,which can effectively detect whale signals in the background of Gaussian noise and ice cracking interference noise.The processing of experimental data in the polar region shows that the traditional detection method and the differential local conditional peak detection algorithm have good detection performance under the spectral level signal-to-noise ratio,but the detection rate of the traditional detection method decreases significantly with the decrease of the dry noise ratio in the polar region environment,while the algorithm in this paper still has a high detection rate,which indicates that the antinoise capability of the algorithm in this paper is better than that of the traditional detection method and has a higher detection accuracy.It has higher detection accuracy and can effectively extract whale signals.Secondly,in terms of feature extraction,four feature extraction techniques are used in this paper,namely Mel frequency cepstrum coefficient based feature extraction algorithm,wavelet packet decomposition of energy features and Gammatone frequency cepstrum coefficient based feature extraction method,and the DWT-MFCC method has been proposed for whale signal feature extraction in polar region environment,and it is compared with the other three feature extraction methods.It is also compared with other three feature extraction methods,which is conducive to better extraction of whale signal features and the extraction of discrimination from the feature domain,laying the foundation for subsequent classification and identification.Finally,by analyzing and processing the whale signals from the measured data,this paper adopts the support vector machine and k-nearest neighbour based on one-dimensional feature input,and the Res Net18 and Le Net5 convolutional neural network models with twodimensional feature input for classification.with better results.Therefore,the automated method proposed in this paper for detecting and classifying whale signals has high detection accuracy and classification accuracy,and can be effectively applied to the detection and classification of whale signals in the Arctic.In conclusion,the automated detection and classification scheme based on the differential local conditional peak detection algorithm and DWT-MFCC feature extraction method proposed in this paper has a certain mechanism in the extraction and identification of whale signals in the Arctic region,which provides a theoretical basis for people to better protect and reasonably use marine biological resources.