Methods for Detecting Novel Geology in Mastcam Multispectral Observations(NPO-51508-1)

Our novelty detection work is implemented in Python using the Keras and TensorFlow machine learning libraries. We created a dataset of Mastcam multispectral images and selected known examples of novel observations to evaluate the novelty detection performance of the compared methods.This is the first work that proposes to train autoencoders using a loss function that explicitly captures spatial properties in the input (e.g., structural similarity index). Additionally, this is the first work to characterize the influence of the autoencoder loss function on the type of novel features that can be detected.