Compact underwater vessel classification improves deployability with auditory inspired features
Towards Deployable Underwater Vessel Classification
SoundMachine Learning
Summary
Classifying underwater vessels by their sounds is challenging, especially for systems that need to be small and efficient. The authors propose a new approach that uses clever ways to represent sound and simple but effective models to recognize different ships from underwater recordings. They found that smaller models can perform well if features and evaluation are carefully designed, but large models like ResNet do not always improve accuracy. The work highlights how choosing the right sound features and evaluation methods matters for making practical, compact underwater acoustic classification systems.
What this means in practice
- •For marine monitoring teams: Classify underwater vessel types efficiently using compact acoustic models suitable for deployment on low-resource systems.
- •For environmental acoustic engineers: Improve automatic identification of ships in underwater environmental data using auditory-inspired sound features and robust evaluation methods.
Authors
Abishek Soti, Thura Pyae Sone, Naqib Ibnul, Htoo Htet Aung, Henry Zhong, Gregory Cohen, Ying Xu
Abstract
We propose a compact underwater acoustic classification framework combining multi-representation feature engineering, temporal statistical pooling, and compact convolutional architectures designed for acoustic time-frequency and cochlear representations. We investigate multiple conventional and auditory-inspired representations and first evaluate lightweight classifiers and Conventional Neural Networks (CNNs) on ShipsEar dataset. On the provided split, a two-layer CNN achieves a macro F1 of 0.9918, while a Radial Basis Function Support Vector Machine (RBF-SVM) reaches 0.9883. However, source-recording provenance cannot be reconstructed, preventing verification of recording-independent generalisation. We therefore evaluate on DeepShip dataset using recording-level partitioning before segmentation. Under this protocol, a 157K-parameter compact CNN achieves a test macro F1 of 0.7226, while an 11.17M-parameter ResNet18 provides no improvement in validation performance under the matched setting. These results demonstrate the importance of representation-aware feature and model design, together with rigorous recording-level evaluation, for classification performance and deployability in compact underwater acoustic systems.