DBM-Based Underwater Acoustic Source Recognition

Underwater acoustic source recognition is a challenging task due to the complex environment of the ocean, which does not have the accurate statistical model and is intractable for analysis. Meanwhile, deep learning technology can learn features from the raw data and it has advantage in extracting the inherent features of signal samples especially under complex environment. Motivated by this, in this paper, an underwater acoustic source recognition based on deep Boltzmann machine (DBM) is proposed. In the proposed method, we choose time-frequency diagrams of the target signal as input units. After non-linear transformation of hidden layers, the data-driven features can be extracted. Finally, a softmax function is used to predict the classification result. Simulation results are shown that the proposed DBM-based method achieves 90% or higher recognition accuracy, which is better than SVM and DBN.