View in EDS

Marine vessels acoustic classification with enhanced machinery and propeller feature extraction, using convolutional neural network.

Saved in:
Bibliographic Details
Title: Marine vessels acoustic classification with enhanced machinery and propeller feature extraction, using convolutional neural network.
Authors: Najamuddin1 (AUTHOR) najamuddin@graduate.utm.my, Sheikh, Usman Ullah1 (AUTHOR) usman@fke.utm.my, Sha'ameri, Ahmad Zuri1 (AUTHOR)
Source: Measurement (02632241). Jan2026:Part A, Vol. 257, pN.PAG-N.PAG. 1p.
Subjects: Convolutional neural networks, Feature extraction, Real-time computing, Underwater acoustics, Ships, Deep learning, Acoustic signal processing
Abstract: Underwater acoustic target detection and classification remain challenging due to the highly dynamic nature of the underwater environment and the adaptive behaviors of targets. Recent progress in data processing and deep learning has driven the development of new techniques aimed at improving the accuracy of detecting and classifying underwater acoustic events. This paper presents a novel preprocessing approach for enhancing weak machinery and propeller signatures. First, the machinery spectrum is estimated using Coherently Averaged Power Spectrum Estimation (CAPSE) and transformed into Low Frequency Analysis and Recording (LOFAR) grams. Simultaneously, Detection of Envelope Modulation on Noise (DEMON) grams are generated using cyclostationary analysis. Key features are automatically extracted from both LOFAR and DEMON grams and concatenated using a customized Convolutional Neural Network (CNN) architecture. The fusion of these features enables the learning of low-level spectral representations, which are subsequently used for marine vessel classification. Experimental results demonstrated that integration of weak machinery signatures extracted using CAPSE and propeller modulation features yields excellent classification performance on publicly available datasets. The proposed model achieves classification accuracies of 98.59% and 97.88% on the ShipsEar and DeepShip datasets, respectively, matching the performance of the best-reported models while maintaining a significantly simpler design. The total number of trained weights is 1.7 million, which is 17.66 times fewer than the best reported classifier. For most vessel types, the model maintains an accuracy above 70% at 5 dB SNR, and demonstrated robust performance with an average accuracy of 60% at 0 dB. This reduction in complexity and training parameters improves the model's efficiency, making it highly suitable for real-time applications without sacrificing classification accuracy. • Novel use of cyclostationary analysis and CAPSE to process marine vessel noise. • DEMON and LOFAR used to extract features from vessel machinery and propeller noise. • Proposed fusion method shows improved results over top models on DeepShip and ShipsEar. [Display omitted] [ABSTRACT FROM AUTHOR]
Copyright of Measurement (02632241) is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Underwater acoustic target detection and classification remain challenging due to the highly dynamic nature of the underwater environment and the adaptive behaviors of targets. Recent progress in data processing and deep learning has driven the development of new techniques aimed at improving the accuracy of detecting and classifying underwater acoustic events. This paper presents a novel preprocessing approach for enhancing weak machinery and propeller signatures. First, the machinery spectrum is estimated using Coherently Averaged Power Spectrum Estimation (CAPSE) and transformed into Low Frequency Analysis and Recording (LOFAR) grams. Simultaneously, Detection of Envelope Modulation on Noise (DEMON) grams are generated using cyclostationary analysis. Key features are automatically extracted from both LOFAR and DEMON grams and concatenated using a customized Convolutional Neural Network (CNN) architecture. The fusion of these features enables the learning of low-level spectral representations, which are subsequently used for marine vessel classification. Experimental results demonstrated that integration of weak machinery signatures extracted using CAPSE and propeller modulation features yields excellent classification performance on publicly available datasets. The proposed model achieves classification accuracies of 98.59% and 97.88% on the ShipsEar and DeepShip datasets, respectively, matching the performance of the best-reported models while maintaining a significantly simpler design. The total number of trained weights is 1.7 million, which is 17.66 times fewer than the best reported classifier. For most vessel types, the model maintains an accuracy above 70% at 5 dB SNR, and demonstrated robust performance with an average accuracy of 60% at 0 dB. This reduction in complexity and training parameters improves the model's efficiency, making it highly suitable for real-time applications without sacrificing classification accuracy. • Novel use of cyclostationary analysis and CAPSE to process marine vessel noise. • DEMON and LOFAR used to extract features from vessel machinery and propeller noise. • Proposed fusion method shows improved results over top models on DeepShip and ShipsEar. [Display omitted] [ABSTRACT FROM AUTHOR]
ISSN:02632241
DOI:10.1016/j.measurement.2025.118494