OCEANS-18

Design and prototyping of a CubeSat-based system to detect undetected vessels through satellite imagery and onboard sensing.

Aerial maritime monitoring view over the Port of Singapore

Overview

Maritime surveillance systems rely heavily on radar and transponder-based tracking, leaving non-cooperative vessels, such as refugee boats, undetected. That creates blind spots in open-sea monitoring and contributes to preventable loss of life.

The goal of OCEANS-18 was to design and prototype a low-cost satellite-based detection system capable of identifying vessels not present in maritime tracking systems, extracting geolocation data from imagery, and supporting cross-referencing with existing maritime databases.

Full Research Paper: IEEE Xplore

System Architecture & Technical Approach

The project was structured as a full-stack aerospace system, combining space, detection, and ground segments.

  1. Space Segment 3U CubeSat structure, optical payload for Earth observation, onboard computing for data acquisition, and a sensor suite including GPS, IMU, barometer, and thermometers.
  2. Detection Pipeline Image acquisition, vessel detection using computer vision, coordinate extraction, and cross-referencing with maritime traffic databases.
  3. Ground Segment Post-processing of collected data and integration with external datasets for validation.

Prototype Development & Testing

A key aspect of the project was the physical development and validation of a working prototype.

CubeSat Prototype

We designed and assembled a 3U CubeSat engineering model including a custom nylon frame, dual onboard computing system, integrated sensor stack, and a power system with batteries and solar panels.

Stratospheric Balloon Test

To simulate near-space conditions, the CubeSat was launched via a stratospheric balloon at 20–40 km altitude and tested for image acquisition, sensor data collection, system stability, and power management.

This helped validate hardware performance under realistic conditions, the data collection pipeline, and the integration between subsystems.

Expected Outputs from Testing

  • Captured images with associated metadata such as position and orientation
  • Real-time telemetry transmission
  • Post-flight data recovery and analysis
Balloon flight setup for the CubeSat deployment test
The balloon deployment setup used to simulate near-space conditions and validate hardware performance, data collection, and subsystem integration.

Detection System & Algorithm Design

The detection system was designed to operate on low-resolution satellite imagery, introducing unique constraints.

Core approach:

  • Object detection using deep learning, specifically Faster R-CNN
  • Transfer learning applied to maritime datasets
  • Multi-band image processing using RGB and infrared when available

Key challenges:

  • Low resolution and limited pixel representation of vessels
  • Environmental variability from weather, lighting, and sea conditions
  • False positives from sea disturbances

Design decisions:

  • Prioritized detection over classification in early stages
  • Focused on robustness across varied atmospheric conditions
  • Leveraged diverse training datasets for generalization
Samples from the MASATI dataset showing single-boat, sea, and multiple-boat images
Samples from the MASATI dataset; among the other labels, there are single boat images (first row), sea images (second row), and multiple-boat images (third row).

Hardware Design Considerations

Payload

An optical camera with 2048×2048 resolution was designed for visible-spectrum acquisition, balancing cost against sensing capability.

Onboard Computing

A dual Raspberry Pi architecture separated image acquisition from sensor data management.

Data Handling

Images were stored onboard because of bandwidth limitations, while metadata was transmitted in real time and post-processing happened at the ground station.

Onboard system diagram with Raspberry Pi and sensors
The onboard architecture diagram shows how the CubeSat splits image acquisition, memory handling, telemetry, and environmental sensing across the Raspberry Pi-based system.

Key Engineering Trade-offs

  1. Cost vs Resolution Low-cost hardware limits detection precision.
  2. Latency vs Accuracy Real-time processing trades off against post-processing reliability.
  3. Coverage vs Infrastructure Satellite scalability must be balanced with deployment cost.

These trade-offs directly influenced system architecture and development priorities.

Results & Learnings

Technical outcomes included the successful design and assembly of the CubeSat prototype, validation of the data acquisition pipeline, and feasibility of vessel detection from optical imagery.

System-level learnings included the importance of latency in real-time tracking scenarios, the sensitivity of detection to environmental conditions, and the need for a scalable constellation for meaningful coverage.

Product & Strategic Considerations

While primarily technical, the project surfaced clear product insights: the real value lies in decision support, integration with existing maritime systems is critical, and scalability through a constellation model is required for meaningful impact.

This reframed the system as a maritime intelligence platform rather than just a sensing tool.

Future Development

  • Onboard real-time processing with edge AI
  • Multi-satellite constellation deployment
  • Improved sensing with infrared and higher resolution
  • Predictive vessel tracking models