Real-Time Acoustic Signal Classification Using RNN for Underwater Cutting Process Monitoring and Situational Awareness

• Published in: Han-guk haeyang gonghak hoeji (Online) Vol. 39; no. 1; pp. 92 - 102
• Main Authors: Pyo, Seunghyun, Yeu, Tae-Kyeong, Lee, Yeongjun, Han, Jong-Boo, Cho, Yujin, Park, Daegil
• Format: Journal Article
• Language: English
• Published: The Korean Society of Ocean Engineers 01.02.2025; 한국해양공학회
• Subjects: acoustic feature extraction; audio classification; long short term memory (lstm); recurrent neural network (rnn); underwater operation; 해양공학
• ISSN: 1225-0767; 2287-6715
• DOI: 10.26748/KSOE.2024.069

Seabed crushing, a critical underwater operation for mineral resource extraction and infrastructure construction, necessitates acoustic monitoring due to the limited visibility caused by debris generated during the process. This study proposes an acoustic classification method to enable real-time monitoring of underwater robotic operations. Given the acoustic characteristics of crushing operations, which predominantly manifest in the low-frequency band, acoustic features were extracted and processed using a deep learning model to classify the operational states into four categories: idling, cutting, hard cutting, and base. The model was developed using a recurrent neural network (RNN), which is particularly suited for real-time time-series data processing. The classification performance of long short-term memory (LSTM) networks and standard RNN models was systematically evaluated. Training on a dataset collected from crushing operations conducted on land, the LSTM model achieved an accuracy of 89%, outperforming the RNN, which achieved 84%. Furthermore, real-time operational state prediction was performed at a speed of 10 Hz, demonstrating high accuracy. These findings indicate that the proposed method effectively enables real-time classification of seabed crushing operations, thereby enhancing the safety and efficiency of remote underwater operations.