Authors:
Pedro Guedes
1
;
José Franco Amaral
1
;
Thiago Carvalho
2
;
1
and
Pedro Coelho
1
Affiliations:
1
FEN/UERJ, Rio de Janeiro State University, Rio de Janeiro, Brazil
;
2
Electrical Engineering Department, Pontifical Catholic University of Rio de Janeiro, Rio de Janeiro, Brazil
Keyword(s):
Neural Networks, Signal Processing, Wavelet Transforms, Underwater Signals, Convolutional Neural Networks.
Abstract:
The identification of underwater sound patterns has become an area of great relevance, both in marine biology, for studying species, and in the identification of ships. However, the significant presence of noise in the underwater environment poses a technical challenge for the accurate classification of these signals. This work proposes the use of signal analysis techniques, such as Mel Frequency Cepstral Coefficients (MFCCs) and Wavelet Transform, combined with Convolutional Neural Networks (CNNs), for classifying ship audio captured in a real-world environment strongly influenced by its surroundings. The developed models achieved a better accuracy in signal classification, demonstrating robustness in the face of adverse underwater conditions. The results indicate the effectiveness of the proposed approach, contributing to advances in the application of neural network techniques to underwater sound signals.