Design and Optimization of 1D-CNN for Spectrum Recognition of Underwater Targets

Abstract

Due to the increasing demand for modern underwater detection, underwater acoustic target recognition (UATR) has always been exploring. By demodulating the underwater target radiated noise (UTRN) received by the passive sonar, the discrete line spectrums reflecting the physical characteristics of the ship can be obtained. In the article, a one-dimensional convolutional neural network (1D-CNN) was proposed to recognize the line spectrums of Detection of Envelope Modulation on Noise (DEMON) spectrum of UTRN. Datasets with different Doppler shifts and signal-to-noise ratios (SNRs) were designed to ensure that the 1D-CNN has an outstanding generalization ability by combining the training and the evaluation process. By exploring a design method of the Field Programmable Gate Array (FPGA)-based CNN model, the DEMON spectrum recognition algorithm based on 1D-CNN was optimized and mapped onto the FPGA embedded platform, which greatly improved the calculation speed while ensuring accuracy. Experimental results show that it is suitable for practical engineering applications, and the recognition process is robust and real-time.