Inspiration
With 4.6 million fishing vessels worldwide, many operate beyond oversight by turning off transponders or falsifying data. We wanted to help expose these “dark vessels” using modern computer vision tools.
What it does
Know Your Vessel detects ships from satellite imagery, matches them to nearby AIS trajectories, and trains a contrastive learning model to recognize vessels based on their visual and motion signatures, even when transponders are off. This enables us to identify repeat offenders or suspicious vessels across time and locations.
How we built it
We combined object detection, geospatial projection, and contrastive learning. Using a pretrained vessel detection model, we localized ships, mapped detections to lat/long coordinates, matched them to AIS data, and trained embeddings that distinguish known vessels from untracked ones.
Challenges we ran into
- Managing GPU constraints during large-scale inference
- Matching AIS data under noisy or missing conditions
Accomplishments that we're proud of
We built a functional data pipeline that performs vessel detection and AIS correlation, setting the groundwork for future identity tracking. We also gained valuable experience handling large satellite datasets and integrating multiple ML components.
What we learned
We learned how challenging it is to make geospatial and temporal data align cleanly, the importance of preprocessing for real-world data noise, and how to debug and optimize GPU-based inference pipelines.
What's next for Know Your Vessel
We plan to extend to multi-frame video tracking, improve rotation-invariant feature learning, and build an interactive analyst UI for exploring similar vessel trajectories and detecting illicit behavior patterns.
Log in or sign up for Devpost to join the conversation.