Intelligent Classification and Recognition of Acoustic Targets Based on Semi-tensor Product Deep Neural Network
Shilei Ma, Haiyan Wang, Xiaohong Shen, Xin Wang · OCEANS 2019 - Marseille · 2019
Traditional acoustic target recognition is mainly based on artificial feature construction. In many cases, there are difficulties in feature construction and low recognition rate. Referring to computer vision technology, this paper proposes a method of acoustic target classification based on semi-tensor product deep neural network. First, the acoustic signal is transformed into Lofargram. Then a semi-tensor product deep neural network model (SPNN) is established. After that the parameters of the SPNN are determined by actual data. Finally, the classification and recognition of sound source targets are realized. Moreover, the recognition accuracy is much higher than that of traditional manual feature extraction and classification by support vector machine (SVM). The recognition rate of underwater target is higher than that of convolution neural network (CNN). The accuracy of air sonar target and CNN is similar, but the training speed of network is much faster.