Research

Method For Underwater Acoustic Signal Classification Using Convolutional Neural Network Combined With Discrete Wavelet Transform

 2023.5.5.

The respected Comrade Kim Jong Un said:

"We should achieve one miracle after another in surpassing the cutting edge, confident that we can beat the world in general science and technology in the near future."

The detection and classification of underwater targets such as fish are one of the major tasks of the underwater acoustic signal processing and very important for scientific, fisheries and ocean engineering and economic fields. The convolutional neural network (CNN) combined with the discrete wavelet transform (DWT) (namely CNN DWT) not only reduces the data processing dimension of signals and the computational costs of the signal analysis, but also improves the performance of target detection and classification.

In order to effectively classify the underwater acoustic signal and increase the accuracy of target detection, researchers from the Institute of Electronic Material, High-tech Research and Development Centre proposed a deep learning method to classify the underwater acoustic signal using CNN combined with DWT, and they studied the work to evaluate the new method for underwater acoustic signal classification compared with classical methods through the simulation experiment.

In the past, the artificial neural networks used for the classification of underwater signals are the shallow ones such as the back-propagation neural network (BPNN). BPNNs are difficult to optimize the network, the generalization ability and accuracy of classification are low in ambient noise, BPNNs converge to local minima, and BPNNs have the limitation that the convergence speed is slow.

Recently, deep neural networks (DNNs) have developed rapidly worldwide and have achieved success in a wide range of tasks. DNNs have the advantage that they barely converge to local minima and their generalization ability is high.

Underwater acoustic signals radiated by underwater targets such as a dolphin are often time-variant, random and non-stationary. As well, because underwater acoustic signals have very low signal-to-noise ratio (SNR), extracting the feature components becomes difficult and the applicability of information drops down. Discrete wavelet analysis is an effective tool for noise elimination and feature extraction and is a powerful mathematic tool in the processing of non-stationary random signal such as the underwater acoustic signal.

A new method of underwater acoustic signal classification is based on CNN combined with DWT. This method consists of steps for white noise elimination, imaging, data augmentation and classification of underwater acoustic signal (Fig. 1). White noise elimination step is the step to remove white noise based on DWT. Imaging step is the one to get an image based on spectrogram of discrete wavelet coefficients for an underwater acoustic signal after removing white noise. Data augmentation step is the one to augment the data based on imaging of underwater acoustic signal. Classification step is the one to classify the underwater acoustic signals using CNN.

Block diagram
Fig. 1. Block diagram of new method for underwater acoustic signal classification

The architecture of new CNN to classify the underwater acoustic signal using in classification step is as Fig. 2. New CNN consists of four convolution blocks, an input layer, a dropout layer, a fully connected layer, a soft max layer and an output layer, and it has 21 layers. A convolution block consists of a convolutional layer, a batch normalization layer, a ReLU layer and a max pooling layer.

Architecture of new CNN
Fig. 2. Architecture of new CNN using for underwater acoustic signal classification

Comparing the new method to the classical methods, the experimental results revealed a substantial increment in classification accuracy and noise robustness. And, the learning curves showed that the proposed CNN had no over-fitting problem, its generalization ability was high and it improved the classification accuracy and convergence of underwater acoustic signals.

We can increase the accuracy of underwater acoustic signal classification and the detection probability of targets by introducing the underwater acoustic instruments such as the fish detector and the net observe implement.

With the classification results and the performance metrics using CNN combined with DWT for underwater acoustic signal classification.

Underwater acoustic signal classification
Fig. 3. Underwater acoustic signal classification using DWT and CNN

The result has been published "Method for Underwater Acoustic Signal Classification Using Convolutional Neural Network Combined with Discrete Wavelet Transform" (https://doi.org/10.1142/S0219691320500927) at International Journal of Wavelet, Multi-resolution and Information Processing (Vol.19, No.4, 2021, 2050092) in 2021.