Development of Rule-Based Autonomous Spectral Analysis Techniques for Planetary Surfaces: Preliminary Results Using Lab Spectra

Many different regression schemes have been used for autonomous identification of minerals from reflectance and emittance spectra. Most existing analysis procedures focus on direct examination of individual spectra or ad hoc procedures to identify specific types of minerals. These procedures are constrained by limitations of the spectral library used as a training set and are inadequate to deal with the enormous volume of data from current and planned missions. One promising alternative to regression schemes is 'expert systems' or rule-based systems, which is one of the oldest AI technologies. A simple rule-based system consists of three components: a list of facts, a set of rules, and a 'production rule interpreter'. The rule-based approach has several advantages over other classification schemes, including the ability to take advantage of the knowledge of human experts, simple problem diagnosis methods, and a `trace' facility to explain reasoning. Because of these advantages development of a rule-based system for autonomous spectral identification on planetary surfaces is underway. The objectives of this project include characterization of spectral bands for selected minerals in laboratory and field spectra and development of algorithms for automated spectral identification of key minerals in visible/NIR and mid-IR spectral ranges. A variety of spectral issues complicate the identification of minerals regardless of whether a human user or an automated system is involved in the processing. These include pure minerals versus mixtures and natural samples, the grain size and texture of minerals, and cation substitutions in the mineral structure. The advantage of the rule-based system is that it is not based on spectral libraries, but on spectral features. Preliminary efforts so far have enabled the autonomous identification of a few minerals based on the presence of multiple spectral features at wavelengths within a defined range using only a band center. Work currently underway includes specifying a band width as well as band center and appears promising for identification of a broader number of minerals.