Marine Life & Vessel Detection

This project develops machine learning systems for detecting, identifying, and classifying both marine life and maritime vessels from acoustic data collected in coastal and offshore environments. The two problems share a common thread: the ocean is full of overlapping sound sources, and separating biological signals from vessel noise — and from each other — requires models that understand the acoustic signatures of each.

The work is a collaboration between UNCW's Center for Marine Science and researchers at the University of South Carolina Beaufort, bringing together marine biologists and remote sensing scientists alongside the machine learning work.

Acoustic recordings and marine traffic data are collected directly by the marine biology and remote sensing teams at UNCW CMS and USC Beaufort — from real underwater environments, not simulated or archival sources. The data captures the full complexity of coastal soundscapes: fish vocalizations, vessel engine signatures, ambient noise, and the interference between them.

This access to field-collected, domain-expert-curated data is central to the project — the ML models are being built against the actual conditions under which they will eventually need to operate.

Maritime vessels produce characteristic acoustic signatures driven by engine type, propeller design, hull size, and speed — features that a trained model can use to detect, classify, and track vessel activity. Coupled with marine traffic data, acoustic vessel detection enables a richer understanding of human activity in coastal waters: where vessels operate, how frequently, and what types are present.

Beyond the classification problem itself, vessel noise is one of the primary sources of interference in marine bioacoustics. Understanding vessel signatures is therefore also a prerequisite for building more robust marine life detectors — separating the two problems helps solve both.