Deep Mars: CNN Classification of Mars Imagery for the PDS Imaging Atlas

NASA has acquired more than 22 million images from the planet Mars. To help users find images of interest, we developed a content-based search capability for Mars rover surface images and Mars orbital images. We started with the AlexNet convolutional neural network, which was trained on Earth images, and used transfer learning to adapt the network for use with Mars images. We report on our deployment of these classifiers within the PDS Imaging Atlas, a publicly accessible web interface, to enable the first content-based image search for NASA’s Mars images.