Acoustic Data Processing Of Underwater Glider And Its Application In Marine Ambient Noise

As the only form of energy that can be transmitted over long distances in the ocean,sound waves are the only effective radiant energy in underwater target detection,underwater communication and navigation.Marine ambient noise is the inherent sound field in the marine,which contains rich environmental information such as sea surface wind speed and precipitation.It is also a kind of interference background field in the underwater acoustic channel,which affects the acoustic communication capability of underwater acoustic equipment.Therefore,long-term and large-scale continuous observation of marine ambient noise has important scientific significance and engineering practical value.Underwater glider is a new type of underwater unmanned submersible vehicle.It has the advantages of large range of activities,low power consumption and strong maneuverability in the observation of marine environmental noise.However,various self-noises of underwater gliders will affect the observation of marine ambient noise as interference sources.Therefore,this paper takes the acoustic data measured by the "Haiyan-L" underwater glider in the southeastern part of the Indian Marine as the research object,and mainly does the following research:Improve the traditional Le Net-5 neural network to realize the classification of underwater glider self-noise.Firstly,the composition of self-noise and its time-frequency domain characteristics were analyzed;Aiming at the characteristics of different energy characteristic information of the self-noise signal through the different frequency bands of the Mel filter bank,a total of 5580 pieces of various self-noise data were intercepted,and the logarithmic Mel spectrogram features were extracted first,and then converted into grayscale images as the input of the model.Through many experiments,we continuously improved the neural network structure,adjusted the parameter settings of the model,and built a new neural network model:the number of convolutional layers is 3 layers,and the depth of each layer is 16,32,and 64 5×5 convolution kernel;and use the Re LU activation function after each convolutional layer and fully connected layer;use the Adam gradient descent algorithm to update the weight parameters of the neural network,the initial learning rate is 0.001,and the descent factor is set to 0.3.At this time,the highest accuracy rate of the model can reach 96.36% after 50 iterations,but there is overfitting.In view of the over-fitting phenomenon of the model,the dropout operation is added to the fully connected layer,which obviously eliminates the phenomenon of classification accuracy curve oscillation.After 100 rounds of iterations,compared with the classification accuracy of 82.47% before improvement,the improved model finally achieved a classification accuracy of 98.06%.After classifying the self-noise of the underwater glider,the marine ambient noise data without the influence of self-noise is intercepted from the observed original acoustic data,and the NCEP sea surface wind speed reanalysis data.Firstly,the time-frequency of the marine ambient noise during the entire observation period is analyzed domain characteristics;since the depth of each ambient noise data segment changes,the relationship between the ambient noise spectral level and the depth is analyzed;finally,the correlation coefficient between the ambient noise spectral level and the wind speed is quantitatively calculated,and the two are compared linear fitting analysis,the following conclusions are obtained: the ambient noise spectral level has a strong consistency with the change trend of the sea surface wind speed,that is,the sea surface wind speed plays a leading role in the ambient noise spectral level;the marine ambient noise spectral level is less affected by the depth change.When analyzing the wind-related characteristics of marine ambient noise observed by underwater glider,the influence of depth changes can be ignored;in the frequency band less than 4 k Hz,it is affected by the noise of the conductivity temperature depth(CTD)pump mounted on the underwater glider,and the correlation coefficient is below 0.2,and the correlation coefficient reaches a high correlation of about 0.8 in the 6-10 k Hz frequency band;in the 5-20 k Hz frequency band,the ambient noise spectral level and the sea surface wind speed fitting curve have a linear function relationship.The spectral level of ambient noise increases with the increase of wind speed,and gradually decreases with the increase of central frequency,and the average frequency attenuation is about5 d B.