Adaptive ship-radiated noise recognition with learnable fine-grained wavelet transform (2306.01002v2)

Abstract: Analyzing the ocean acoustic environment is a tricky task. Background noise and variable channel transmission environment make it complicated to implement accurate ship-radiated noise recognition. Existing recognition systems are weak in addressing the variable underwater environment, thus leading to disappointing performance in practical application. In order to keep the recognition system robust in various underwater environments, this work proposes an adaptive generalized recognition system - AGNet (Adaptive Generalized Network). By converting fixed wavelet parameters into fine-grained learnable parameters, AGNet learns the characteristics of underwater sound at different frequencies. Its flexible and fine-grained design is conducive to capturing more background acoustic information (e.g., background noise, underwater transmission channel). To utilize the implicit information in wavelet spectrograms, AGNet adopts the convolutional neural network with parallel convolution attention modules as the classifier. Experiments reveal that our AGNet outperforms all baseline methods on several underwater acoustic datasets, and AGNet could benefit more from transfer learning. Moreover, AGNet shows robust performance against various interference factors.

• The paper introduces AGNet, which uses learnable fine-grained wavelet parameters to dynamically extract features from complex underwater ship noise.
• It integrates parallel convolutional attention modules to focus on essential time-frequency information, improving recognition accuracy.
• Experimental results demonstrate high accuracy across datasets, showcasing AGNet’s robustness against low SNR and environmental noise.

Adaptive Ship-Radiated Noise Recognition Using Fine-Grained Wavelet Transformations

The intricacies of underwater acoustic environments, characterized by significant background noise and fluctuating transmission conditions, challenge the efficacy of traditional ship-radiated noise recognition systems. Existing systems often lack robustness in various underwater settings, thus underperforming in practical applications. In response, Xie et al. propose AGNet (Adaptive Generalized Network), a novel recognition framework that leverages a learnable fine-grained wavelet transform to better adapt to these dynamic conditions.

Core Contributions

The paper introduces a pivotal advancement by converting fixed wavelet parameters into fine-grained learnable vectors, allowing AGNet to adaptively interpret the acoustic signatures across different frequencies. The adaptive wavelet transform significantly enhances feature extraction by automatically updating wavelet parameters such as order, bandwidth, and center frequency in a data-driven fashion. This allows for more precise capturing of underwater sound characteristics, including complex signals influenced by noise and varying channel properties.

Furthermore, the paper advances the application of parallel convolutional attention modules within AGNet's convolutional neural network (CNN)-based classifier. This mechanism focuses computational resources on valid time-frequency domain information within wavelet spectrograms, optimizing recognition accuracy.

Experimental Results

AGNet demonstrates superior performance compared to all baselines across multiple underwater acoustic datasets. On the Shipsear dataset, AGNet achieves a recognition accuracy of 85.48%, while attaining 77.09% on DeepShip and 95.76% on the DTIL dataset. These results underscore AGNet's enhanced generalization capability and adaptability through low SNR (Signal-to-Noise Ratio) scenarios, facilitated by the adaptive wavelet transformation. The system showed resilience to environmental interference factors, with considerable tolerance to additive colored noise and variations in cutoff frequency.

Implications and Future Directions

AGNet's implementation marks a strategic shift in underwater acoustics, showcasing the potential of adaptive learning parameters in marine signal processing. Its end-to-end architecture reduces the complexity of real-world deployment by automating parameter tuning and minimizing module integration challenges. The integration of transfer learning opens novel avenues for leveraging large-scale audio datasets to augment recognition frameworks for underwater environments.

This work lays a foundation for future research concerned with adaptive parameter learning in complex acoustic settings. Expanding the learnability of parameters within feature extraction could further optimize model performance. Future models might explore integrating additional contextual data to refine spectrogram analysis further, aiming to improve accuracy and reduce misclassification in ambiguous cases.

Overall, this paper represents a significant contribution to the domain of marine acoustics, offering a robust and scalable solution to ship noise recognition. Its approach may inspire further investigations where adaptive techniques are necessary to address the challenges posed by volatile and information-rich environments.