Open AccessACOUSTIC SIGNAL ANALYSIS AND CLASSIFICATION BASED ON NEURAL NETWORK ALGORITHMS
Author(s) -
Said Jamal,
Yahya Benremdane,
LAKZIZ JAWAD
Publication year - 2022
Publication title -
journal of marine technology and environment
Language(s) - English
Resource type - Journals
eISSN - 2501-8795
pISSN - 1844-6116
DOI - 10.53464/jmte.01.2022.08
Subject(s) - computer science , noise (video) , frequency domain , time domain , signal processing , artificial neural network , filter (signal processing) , artificial intelligence , signal (programming language) , narrowband , process (computing) , identification (biology) , pattern recognition (psychology) , algorithm , computer vision , radar , telecommunications , operating system , botany , image (mathematics) , biology , programming language
"This paper presents the results of an innovative approach in the underwater domain of research related to the identification, classification and recognition of maritime targets using acoustic data processed. The “Acoustic Signature” is specific to each target type; it is usually produced by the vibration of the propulsion system of surface vessels caused by their radiation into the water. Therefore, the processing of the frequencies generated by these vibrations is essential for the analysis and the classification of different target type. The purpose of this study is to build an alternative method to identify and classify targets with passive sonars using the TPWS (Two - Pass Split - Windows) filter. In this process, the signal generated by the target is processed in time frequency domain. Then a TPSW algorithm is applied in the frequency domain to reduce the background noise and enhance the frequency lines of the target noise. Finally, an artificial intelligence model is applied to classify targets, taking as inputs the narrowband and the broadband analysis. This classification is based on deep learning process, relied on the training, validation, and test phases in order to enhance the accuracy and reduce the loss. Our results showed that the suggested method is accurate (appx 83.5% SNR = 0db), depending essentially on the signal/noise ratio. "