Mineral emittance spectra: Clustering and classification using self-organizing maps

Remote sensing of terrestial and planetary surfaces in the infrared for the purpose of identifying minerals and their distribution is an ongoing activity. The huge amount of spectral data currently available and expected from future space missions presents a challenge to the scientist determine to extract useful scientific information from this data. Automated methods to facilitate this process are clearly needed to complete these investigations in a timely manner. Organizing this data into meaningful chunks is a crucial first step of this analysis that can be be addressed with clustering techniques. We have developed an automated clustering and classification scheme based on kohonen self-organizing maps (SOM). The SOM is a type of unsupervised neural network widely used to cluster data and identify anomalies. A modification of the davies-bouldin (DB) cluster validation index which incorporates the measurement uncertainties of the SOM training data is used to determine the optimal number of clusters. This scheme was trained and tested with the mineral spectral libraries prepared at Arizona State University (ASU) and John Hopkins University (JHU) whose samples have been hierarchically labeled with Class, Subclass and Group names. These names are used to measure the spectral purity of the derived clusters and the accuracy of classification. We describe in detail the SOM scheme itself, the performance measures, and testing procedure. The test results demonstrate that the SOM can be a useful component in autonomous systems designed to identify minerals from reflectance or emissivity measurements in the thermal infrared.