The Computation of Generalized Embeddings for Underwater Acoustic Target Recognition using Contrastive Learning

TL;DR

This paper tackles the scarcity of labeled data in underwater acoustic target recognition by applying unsupervised contrastive learning (CL) using VICReg on unlabeled single-hydrophone data. A Conformer-based encoder learns robust, generalized embeddings that transfer to downstream ship-type and marine mammal classification tasks, achieving competitive performance against supervised CL baselines and demonstrating cross-dataset generalization. The study analyzes loss components, augmentation strategies, embedding dimensions, and data-split effects, offering practical insights for deploying UATR in data-limited environments. Overall, the approach shows strong potential to automate large-scale, label-sparse underwater acoustic analysis across diverse tasks.

Abstract

The increasing level of sound pollution in marine environments poses an increased threat to ocean health, making it crucial to monitor underwater noise. By monitoring this noise, the sources responsible for this pollution can be mapped. Monitoring is performed by passively listening to these sounds. This generates a large amount of data records, capturing a mix of sound sources such as ship activities and marine mammal vocalizations. Although machine learning offers a promising solution for automatic sound classification, current state-of-the-art methods implement supervised learning. This requires a large amount of high-quality labeled data that is not publicly available. In contrast, a massive amount of lower-quality unlabeled data is publicly available, offering the opportunity to explore unsupervised learning techniques. This research explores this possibility by implementing an unsupervised Contrastive Learning approach. Here, a Conformer-based encoder is optimized by the so-called Variance-Invariance-Covariance Regularization loss function on these lower-quality unlabeled data and the translation to the labeled data is made. Through classification tasks involving recognizing ship types and marine mammal vocalizations, our method demonstrates to produce robust and generalized embeddings. This shows to potential of unsupervised methods for various automatic underwater acoustic analysis tasks.

Figures (8)

• Figure 1: The proposed architecture for the unsupervised framework. The framework starts with a Mel spectrogram, followed by four Conformer Blocks (indicated in black), one flatten layer combined with a linear layer (indicated in brown), and finalized with an Expander (indicated in light brown).
• Figure 2: Visualization of the augmentation family proposed for defining positive samples during training. The top image shows the pipeline for the MixUp strategy and the down image shows the resulting spectrograms for all augmentation functions.
• Figure 3: A single Conformer Block.
• Figure 4: Overview of the experiments.
• Figure 5: Accuracy of supervised baseline models trained on either time-wise split or random split of Deepship.