In a significant technical milestone, we've recently completed our prototype ship detection AI model. This ship detector is the foundation of several applications of high revisit rate, persistent Earth observation, including stopping maritime piracy and providing maritime border surveillance. A sample image is provided above, with several ships detected by the ship detector outlined in blue.
Satellite images used are 3m resolution PlanetScope images provided by Planet. This ship detector was achieved with a Convolutional Neural Network (CNN), trained with only 13 images and 1000 data points. It achieved approximately 75% accuracy. We're proud of this as it's usually expected that millions of data points are needed to achieve any reasonable accuracy in a CNN.
The next steps for Spiral Blue are to consult with members of industries including maritime, defence, resources and insurance to start commercialising this capability.