Improving the classification of propeller ships using LOFAR and triple loss variational auto encoder

Abstract

This paper presents an underwater signal processing model for the purpose of detecting and classifying propeller ship by the Low Frequency Analysis and Recording (LOFAR) algorithm combined with the Triple loss Variational Auto-Encoder network (TL- VAE). The results of the model have been tested on real data sets of Deepship, and showed better classification accuracy than Convolutional Neural Network (CNN) VGG-19. By replacing FFT with STFT before normalizing by TPSW (Two pass split window) and using the spatial domain probability distribution, the proposed model LOFAR-TL-VAE improved the classification accuracy by 10% even with low signal to noise ratio actual signals.