Enhanced Spectral Anomaly Detection in Martian Regolith via Deep Variational Autoencoder (DVAE) and Hierarchical Clustering

**Abstract:** This paper introduces a novel methodology for automated spectral anomaly detection in Martian regolith data acquired by ChemCam/SuperCam instruments. Leveraging recent advancements in deep learning, specifically Deep Variational Autoencoders (DVAE) and hierarchical clustering, we present a system capable of identifying subtle spectral variations indicative of potential mineralogical anomalies with significantly improved detection accuracy and reduced false positives compared to traditional threshold-based methods. The approach is designed for real-time onboard processing by future Martian rovers, enabling autonomous decision-making regarding targeted investigations. This system has the potential to revolutionize the identification of key geological features on Mars, significantly enhancing scientific discovery potential within a 5-10 year timeframe.