The Anthropocene Institute is an organization based in Palo Alto, California, focusing on exploring solutions to tackle climate change. With a goal of solving the climate problem by 2030, the Anthropocene Institute connects investors, entrepreneurs and policymakers with research groups and experts, while assessing research claims and viability.

Marine conservation is a key part of maintaining and healing entire ecosystems and populations that depend on the ocean. Focusing on enforcing and monitoring marine conservation and no-fishing zones, ProtectedSeas, a partner of the Anthropocene Institute, utilizes cameras on their land-based radar systems to capture images of vessels near protected zones to ensure compliance. Using these images, ProtectedSeas is creating ship-identifying AI models and must perform the time-consuming task of hand-labeling thousands of images for model training.

Our Vessel Classifier for Marine Monitor (M2) takes input images and identifies whether a ship is present using machine learning, automating the labeling process, saving time and effort.

The system runs on a website that takes in user images and automatically labels them using a model trained periodically on the vessel dataset. If the model is not at least 90 percent confident with its label, the user is prompted to manually classify said image. The user can also access all the images from that session to override their automated labels. The images and their labels are added to ProtectedSeas’ database to train their ship-identifying models as well as the system itself periodically.

Using our system, training AI is more efficient than ever before, giving ProtectedSeas more time to keep the ocean protected.

Our Vessel Classifier for Marine Monitor (M2) runs on a Flask website containerized in Docker for ease of migration and uses a PyTorch computer vision model for image classification.