ONCship: A Large-Scale Underwater Acoustic Corpus for Ship Recognition

Zhe Cao; Jichao Zhang; Xiao-Lei Zhang

cs.SDOct 14, 2026

ONCship: A Large-Scale Underwater Acoustic Corpus for Ship Recognition

Authors: Zhe Cao, Jichao Zhang, Xiao-Lei Zhang

Organizations: School of Marine Science and Technology, Northwestern Polytechnical University, Xi’an, 710072, China · Research and Development Institute of Northwestern Polytechnical University in Shenzhen, Shenzhen, 518063, China

Abstract

Recent advances in deep learning have significantly improved underwater acoustic vessel recognition. However, existing publicly available datasets are typically limited in scale, hindering large-scale representation learning and downstream task evaluation. To address this limitation, we introduce ONCship, a large-scale underwater acoustic dataset comprising 4521 hours of curated recordings from Ocean Networks Canada (ONC) hydrophones and Automatic Identification System (AIS) data. We evaluate the proposed dataset across three tasks: self-supervised representation learning, vessel-type classification, and MMSI-Level Vessel Classification. Experiments on DeepShip and ShipsEar demonstrate the effectiveness of ONCship for acoustic representation learning, with joint pre-training on ONCship and AudioSet achieving a classification accuracy of 62.45% on DeepShip through linear probing. Furthermore, incorporating ONCship as auxiliary training data improves downstream vessel-type classification performance. Beyond conventional vessel-type classification, we further investigate MMSI-Level Vessel Classification, treating each vessel as a distinct identity. Experimental results show that models fine-tuned with different loss functions achieve Top-1 accuracies exceeding 53% and Top-5 accuracies approaching 69%, demonstrating the feasibility of distinguishing individual vessels through MMSI-Level Vessel Classification. Overall, ONCship provides a valuable resource for underwater acoustic representation learning and vessel recognition, while its rich vessel-level metadata enables further exploration of MMSI-Level Vessel Classification.

Figures & tables

Figure 1: Example log-mel spectrograms for the four most frequent vessel types in ONCship: Tug, Cargo, Passenger, and Towing. Each panel shows one recording.

Dataset	Year	Duration (h)	Label scope	Source
ShipsEar Santos-Domínguez et al. (2016)	2016	2.8	11 vessel types	Vigo, Spain
DeepShip Irfan et al. (2021)	2021	47.1	4 vessel types	ONC, Canada
QiandaoEar22 Du and Hong (2024)	2024	31.4	20 target labels	Qiandao Lake, China
Oceanship Li et al. (2024)	2024	121.2	15 vessel types	ONC, Canada
IARA da Silva et al. (2025)	2025	129.6	34 vessel types	Santos Basin, Brazil
UniqueShip Hashemi et al. (2026)	2026	3437.0	11 vessel types	ONC, Canada

Table 1: Selected underwater acoustic datasets. Durations are approximate totals in hours, including background where available.

Figure 2: ONCship construction pipeline. The five steps collect hydrophone and AIS data, clean and interpolate vessel trajectories, associate trajectories with hydrophones, label 30s time bins, and merge segments before sampling 60 s clips. Sampled clips are subsequently subdivided into 10s clips.

Panel of Figure 3 — Figure 3: Locations of the ONC hydrophone stations and associated AIS receivers. Left: DIIP and DISS. Center: CRIP and CRSS. Right: the Vancouver-area stations and IONA. Base map: OpenStreetMap contributors, available under the Open Database License ( https://www.openstreetmap.org/copyright ).

Station	Lat., Lon. ( ∘ )	Device ID	Recording period	AIS	AIS Lat., Lon. ( ∘ )
DIIP	$54.259,-130.431$	ICLISTENHF1352	2016.03.30–2017.04.06	DISS	$54.263,-130.425$
		ICLISTENHF1352	2017.04.07–2018.10.25
		ICLISTENHF1354	2018.10.26–2019.08.29
		ICLISTENHF1352	2019.08.29–2021.06.02
		ICLISTENHF1293	2021.06.02–2021.10.12
		ICLISTENHF1293	2021.10.12–2022.07.16

Table 2: Complete hydrophone deployment history with the associated AIS receivers. Coordinates are decimal degrees; positive latitude denotes north and negative longitude west. A dash denotes that AIS records were not downloaded.

Figure 4: ONCship distribution. (a) Duration by recording category. (b) Major vessel-type labels among single-ship recordings with type metadata; Others aggregates all remaining labels, including Reserved and Unknown.

Category	Duration (h)
Single-ship, labeled	3106.8
Single-ship, unlabeled	38.4
Multi-ship	929.1
No-ship	7.4
USSLP (AIS not downloaded)	439.3

Table 3: Total duration of each segment category after filtering and clipping.

Metadata label	Clips
Tug	237238
Cargo	191287
Passenger	188183
Towing	182301
Pleasure	117036
Fishing	81616

Table 4: Number of 10 s clips per raw vessel-type metadata label. SAR: search and rescue; HSC: high-speed craft; WIG: wing-in-ground craft; RR18: noncombatant vessels covered by Radio Regulations Resolution No. 18 (AIS ship-type code 59) Guo et al. (2023) .

Model	DeepShip			ShipsEar
Model	NMI	AUC	ACC (%)	NMI	AUC	ACC (%)
EAT-as	0.1877	0.6297	61.75	0.4280	0.7284	70.40
EAT-onc	0.2021	0.6437	61.35	0.3727	0.7127	62.36
EAT-as&onc	0.1993	0.6540	62.45	0.3882	0.7264	76.18

Table 5: Comparison of EAT models pre-trained from scratch on different datasets. NMI, AUC, and ACC denote normalized mutual information, feature-retrieval AUC, and classification accuracy, respectively. The best result in each column is highlighted in bold.

Model	DeepShip		ShipsEar
Model	ACC (%)	Weighted F1 (%)	ACC (%)	Weighted F1 (%)
ResNet18 (Adam)	53.94	49.00	21.40	8.00
ResNet18 (LARS)	57.78	56.00	43.94	42.00
EAT-as	61.75	61.93	70.40	70.85
EAT-onc	61.35	61.13	62.36	61.85
EAT-as&onc	62.45	62.49	76.18	77.25

Table 6: Comparison of downstream classification performance between EAT models and contrastive learning baselines. ACC and Weighted F1 denote classification accuracy and weighted F1-score, respectively. The best reported result in each column is highlighted in bold.

Model	DeepShip			ONCship (Different Hydrophones)			ONCship (Same Hydrophones)
Model	ACC (%)	macro-F1 (%)	macro-Recall (%)	ACC (%)	macro-F1 (%)	macro-Recall (%)	ACC (%)	macro-F1 (%)	macro-Recall (%)
ResNet50	$39.45\pm 7.94$	$30.53\pm 7.03$	$36.78\pm 5.76$	$37.63\pm 6.35$	$30.00\pm 8.02$	$36.31\pm 7.82$	$36.38\pm 6.86$	$28.89\pm 9.31$	$34.93\pm 8.30$
EAT-as	$65.64\pm 5.96$	$64.08\pm 4.82$	$65.38\pm 4.24$	$65.95\pm 5.16$	$64.39\pm 3.52$	$64.92\pm 3.37$	$68.16\pm 3.75$	$65.74\pm 1.75$	$66.74\pm 1.70$
BEATs	$\mathbf{68.11\pm 3.95}$	$\mathbf{65.65\pm 2.64}$	$\mathbf{65.99\pm 2.68}$	$\mathbf{68.85\pm 4.43}$	$\mathbf{66.30\pm 3.25}$	$\mathbf{66.86\pm 2.78}$	$\mathbf{69.41\pm 2.22}$	$\mathbf{67.03\pm 0.25}$	$\mathbf{67.83\pm 1.27}$

Table 7: Fine-tuning performance on DeepShip under different training data configurations. ACC, F1, and Recall are reported as percentages (mean

\pm

standard deviation over three runs). The highest mean within each training configuration is highlighted in bold.

Figure 5: Comparison between conventional vessel-type classification and MMSI-Level Vessel Classification. Conventional vessel-type classification assigns underwater acoustic recordings from different vessels to coarse-grained vessel categories (left), whereas MMSI-Level Vessel Classification aims to identify individual vessels based on their MMSI labels (right). Recordings from the same vessel share a common MMSI label, while different vessels of the same type are treated as distinct identities.

Model	Top-1 ACC	Top-5 ACC	Macro-F1
EAT(CE)	0.5312	0.6957	0.5307
EAT(ArcFace)	0.5320	0.6892	0.5333

Table 8: MMSI-Level Vessel Classification results. EAT(CE) denotes the model trained with cross-entropy loss, while EAT(ArcFace) employs ArcFace loss. The best result in each column is highlighted in bold.

Explore similar work

Sep 12, 2026

UniqueShip: Mitigating Data Leakage in Acoustic Ship Classification Benchmark Datasets

Underwater Acoustic Target Recognition (UATR) of ships is well-suited for machine learning, yet its progress is hindered by the lack of large, diverse, and publicly available labeled datasets. In this work, we introduce UniqueShip, a machine learning-ready benchmark dataset for UATR applications sourced from the open Ocean Networks Canada (ONC) repository. Unlike previous datasets, we explicitly control for "data leakage" between the training and evaluation sets to ensure more reliable and generalizable model evaluation that does not encourage the model to memorize individual ships. We demonstrate that typical, random data partitioning in two prominent UATR datasets leads to falsely optimistic test performance, increasing accuracy by 10-48 percentage points compared to our more careful partitioning. Ablations on UniqueShip further show that doubling the number of unique vessels improves accuracy by 2.4-2.6 percentage points, while doubling total audio duration improves only by 0.8-1.3 points, indicating that vessel diversity should drive dataset curation more than total hours. We provide baselines with convolutional and transformer backbones, and analyze how ship metadata correlates with classification performance, finding that individual vessel characteristics predict classification difficulty far better than distance to the hydrophone alone. Overall, UniqueShip contains 2,460 hours of ship-radiated audio from 4,218 unique vessels (3,437 hours including background). We publish the dataset, code, and easy-to-download splits at uniqueshipdata.org to foster further UATR research.

Connor Hashemi, Trevor Stout, Anthony Hoogs +1

Kitware, Inc., 1712 Route 9, Suite 300, Clifton Park, NY 12065, USA

Audio Classification Data Leakage Classification

Ask PDF

Sep 24, 2026

Towards Deployable Underwater Vessel Classification

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.

Abishek Soti, Thura Pyae Sone, Naqib Ibnul +4

ICNS, Western Sydney University, Sydney, Australia

Passive Acoustic Monitoring Lightweight CNNs Audio Classification

Ask PDF

Jun 27, 2026

Underwater Source Detection and Classification for Signal-based Surveillance: Audio Dataset Curation and Cross-Domain Evaluation

Machine learning for underwater acoustics is constrained by the scarcity of publicly available labeled datasets. In contrast to air-acoustic domains, where large benchmarks enable rapid model development, underwater datasets are typically small and limited in acoustic diversity, restricting robust model training and cross-domain generalization. To help address this gap, we introduce a curated underwater audio dataset derived from an open-source maritime sound archive. The dataset contains over one thousand labeled audio segments across eight biologically and mechanically relevant acoustic classes, providing an additional resource for training models in data-limited underwater environments. Additionally, we establish a lightweight Convolutional Neural Network (CNN) baseline and propose a margin-enhanced loss with feature alignment to mitigate class confusion arising from data imbalance, acoustic similarity, and cross-domain mismatch. While the baseline achieves 96.35% in-domain accuracy, evaluation on ShipsEar reveals substantial domain shift; the proposed feature alignment improve zero-shot ship detection by 42.60%, demonstrating stronger robustness under distribution mismatch. We further release a transparent curation pipeline and reproducible benchmark to support future research on imbalance mitigation, domain adaptation, and data-efficient underwater acoustic classification.

Quoc Thinh Vo, David K. Han

Department of Electrical and Computer Engineering, Drexel University, Philadelphia, PA, USA

Training Data Curation Audio Classification Cross-Dataset Generalization

Ask PDF