In collaboration with Dr. Diana Gentry at NASA Ames, I have been developing methods for spectral data classification as part of the Statistical Classification of Biosignature Information (SCOBI) project. Here, a workflow for feature extraction from reflectance spectra from six public planetary science databases is shown.
Four groups: indicative alive (e.g. biofilm), indicative not alive (e.g. bone), indicative mixed (e.g. snow), and not indicative of life (meteorite), are used.
Specific features, such as number of peaks, can be extracted from each reflectance spectrum. These can be used to represent the spectrum for classification. Classification is performed by multiple different algorithmic approaches, including KNN and CNN (image based classifying).