AIS-annotated Hydrophone Recordings for Vessel Classification

This dataset contains 10-second underwater acoustic recordings labeled with Automatic Identification System (AIS) data, collected from the Belgian part of the North Sea (BPNS). The data were gathered to develop machine learning models for vessel activity classification and distance prediction. Hydrophones were deployed at two stations, Gardencity and Grafton, near busy shipping routes, capturing vessel-generated sounds. AIS data was used to annotate these recordings with vessel position, speed, type, and activity. The dataset includes 27 524 labeled audio segments recorded over 116 days and is split into training, validation, and testing subsets.   This dataset was developed to support research on vessel monitoring using underwater acoustic recordings labeled with AIS data. It was created by researchers from VLIZ (Flanders Marine Institute) and collaborators to improve machine learning models for vessel classification and distance prediction. The dataset consists of 10-second underwater sound recordings, collected at two hydrophone stations, Gardencity and Grafton, located in the Belgian part of the North Sea (BPNS). These stations were strategically placed near major shipping routes to capture vessel-generated noise. Vessel positions, speeds, types, and activities were extracted from AIS-Hub data and linked to each recording, providing labeled data for model training. The recordings were collected over 116 days, resulting in 27 524 labeled audio segments. Each segment is accompanied by metadata, including AIS-derived vessel information and the hydrophone station. The dataset splits (training, validation, and testing) are provided as separate files in the data_split folder to ensure structured and reproducible dataset usage for machine learning applications. This dataset enables research in passive acoustic monitoring, vessel detection, and maritime traffic analysis, offering valuable data for studying human activity at sea.