In the domain of planetary science, novelty detection is quickly gaining attention because of the operational solutions it offers, including annotated data products and downlink prioritization. When detecting novelties in images, autoencoders have shown to have value, both in their predictive properties and their visualizations. In this study, a processing pipeline that supports rapid autoencoder prototyping for novelty detection is presented, along with two autoencoder variants. Models are trained on two planetary datasets: The Moon and Mars. Results show that these networks outperform state-of-the-art networks by over 6% when monitoring the area under the receiver operating characteristic curve. Operational viability of the proposed autoencoder networks is discussed.