ONCship: A Large-Scale Underwater Acoustic Corpus for Ship Recognition
Underwater vessel recognition is limited by scarce training data and small public benchmarks. This paper introduces ONCship, a 4521-hour corpus curated from Ocean Networks Canada hydrophones and Automatic Identification System vessel metadata. It evaluates domain-specific self-supervised pre-training, supplementary supervised fine-tuning, and MMSI-Level individual vessel classification, showing that scale and metadata can improve representations even though downstream gains are architecture-dependent and uneven.
The paper makes a solid empirical contribution by curating a large-scale underwater acoustic dataset and benchmarking it under controlled splits. The finding that joint AudioSet and ONCship pre-training yields the best linear-probing accuracy is well-supported: 'Combining AudioSet and ONCship further improves performance, achieving the highest AUC of 0.6540 and ACC of 62.45% on DeepShip' (Sec 4.3). However, the work is ultimately a dataset and benchmarking paper whose broader impact hinges on whether the curated corpus and processing code will actually be released.
The dataset construction is transparent about avoiding benchmark leakage: 'For self-supervised pre-training, we exclude recordings from the FAE and LSBBL stations used by DeepShip, leaving 3905h of ONCship data' (Sec 1). Keeping three pre-training regimes under the same EAT architecture isolates the effect of data source on representation quality, and the experiments demonstrate consistent gains for joint pre-training on retrieval AUC (0.6540) and linear classification accuracy. The MMSI-Level task is also a useful extension beyond coarse type labels.
The single-ship and no-ship labels rely on heuristic AIS association rules that do not guarantee acoustic purity. The authors note that 'the no-ship label is defined by these AIS-based rules, not by verified absence of all shipping noise' (Sec 3.2) and that 'single-ship and multi-ship refer to the number of retained near-vessel trajectories, not to all vessels within 50km' (Sec 3.2), leaving room for unlabeled distant shipping noise in supposedly clean clips. The vessel-type distribution is also severely imbalanced: 'the number of clips varies by several orders of magnitude across vessel types, resulting in a pronounced long-tailed distribution' (Sec 3.3), yet no class-balancing strategy is used in supervised training. Downstream fine-tuning does not uniformly help: 'ResNet50 exhibits a decline in performance when additional ONCship data are introduced' (Sec 4.4), suggesting that supplementary data can hurt depending on model capacity and acoustic mismatch. Finally, the MMSI-Level benchmark is limited to only 50 vessels and 100 hours, which is too small to substantiate claims about large-scale fine-grained identification.
The frozen-encoder ablations support the claim that ONCship improves transferable representations, but the comparison against LARS contrastive learning confounds architecture and data because the latter uses ResNet18 while the EAT models use a Transformer. The paper acknowledges that pre-training benefits vary across benchmarks, yet it does not explain why underwater-only pre-training improves DeepShip NMI but degrades ShipsEar NMI relative to AudioSet. Comparisons to related datasets are accurate: the description of UniqueShip as containing 'approximately 3437h of audio, including background recordings, and includes recordings associated with 4218 distinct vessels' (Sec 2.1) matches its abstract, which reports 'Overall, UniqueShip contains 2,460 hours of ship-radiated audio from 4,218 unique vessels (3,437 hours including background).' (Hashemi et al., Abstract).
While pre-training hyperparameters are clearly listed, the paper does not state whether ONCship clips, metadata, or processing code will be publicly released. Reproducing the dataset from the raw 20,000-hour ONC archive would require replicating a multi-step pipeline that includes 'trajectory cleaning and interpolation, spatial association, temporal labeling, and segment merging with clip sampling' (Sec 3), and small differences in the 2 km association radius or interpolation rules could substantially alter the final corpus. Until a public release is confirmed, independent reproduction is effectively blocked.
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.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.