Leveraging satellite imagery to determine Oil spills at sea along with AIS data correlations to identify vessel responsible for the spill.
“Marine oil spills destroy aquatic ecosystems, but it is often hard to figure out which ship caused them. This project builds an automated system using satellite pictures and ship-tracking data to spot oil slicks, trace where they came from, and identify the responsible vessel.”
Deconstructed Problem Blueprint
WHAT TO BUILD?
An intelligent geospatial analytics platform comprising an AI-powered satellite image processing pipeline, an ocean drift trajectory simulator (hindcasting/forecasting), and an AIS vessel traffic correlation and scoring engine wrapped in an interactive dashboard.
WHO IS IT FOR?
The National Technical Research Organisation (NTRO), coastal guard units, maritime security agencies, and environmental protection bureaus.
WHY DOES IT EXIST?
Ships illegally dump oil or suffer leaks at sea and sail away undetected because manually cross-referencing vast ocean areas with millions of satellite frames and historical ship logs is painfully slow and complex.
INPUT → PROCESS → OUTPUT ARCHITECTURE PIPELINE
Sentinel-1 Synthetic Aperture Radar (SAR) and Electro-Optical (EO) satellite imagery, historical/synthetic AIS (Automatic Identification System) vessel tracking data, and marine meteorological/oceanographic current data.
Computer vision semantic segmentation (e.g., U-Net) to detect oil slicks, geometric and age profiling, physics-based hydrodynamic drift simulation for backward/forward tracking, and spatio-temporal matching of filtered vessel tracks against the spill origin.
An interactive maritime dashboard displaying oil slick overlays, predicted drift paths, ranked suspect vessel profiles with confidence scores, and automated evidentiary export reports.
CONSTRAINTS
- Must handle large geospatial raster files efficiently without crashing the UI
- SAR image speckle noise and high false-positive rates from natural phenomena like algae or low-wind areas
- AIS data gaps due to intentional transponder turn-offs (dark vessels)
JURY METRICS
- Segmentation accuracy (IoU / Dice coefficient) for oil slick detection
- Distance error between estimated spill origin and actual vessel location
- Processing inference latency per satellite frame
- UI responsiveness and clarity of the suspect ranking explanation
CORE BOTTLENECK
Accurately simulating backward ocean drift (hindcasting) using sparse or noisy current vectors to pinpoint the exact time and space coordinate of the spill origin.
Why You Should Choose This PS (Pros)
- High national security and environmental impact solving an elite agency's (NTRO) real-world challenge
- Visually stunning demo potential with maps, heatmaps, and animated drift trajectories
- Clear modular separation of concerns: CV model, drift simulator, and AIS correlation engine
Challenges & Risks to Watch (Cons)
- Requires heavy geospatial libraries (GDAL, rasterio, Shapely) which can be tricky to configure under hackathon time limits
- Complexity of oceanographic physics models for drift calculation
- Acquiring and preprocessing real SAR imagery can have a steep learning curve