Rockspec Project
Overview
Rockspec is a Java application (JDK 1.4+), the goal of which is to help compare the accuries of various algorithms at classifying rocks by mineral composition based on spectal data. It has been used to comparing the Modified PC algorithm, the Regression Variable Selection algorithm, and other algorithms, as identifiers of carbonate based on visual to near infrared (VNIR) reflectance spectral data (0.4 to 2.5 um). Some of these results are reviewed on this website. It has also been applied to infrared (7.5 to 24.5 um) emission spectral data; some of these results as well are supplied as well.
Rockspec is primarily a tool to help load spectral data in various formats, preprocess the data in ways that are common, and apply classification algorithms to the data, generating reports that can be compared across algorithms. The current version of Rockspec (and sample data files) can be downloaded from this website using the links to the left.
One example of why NASA scientists are interested in mineral class idenficationof rocks based on spectra is the they hope such classification techniques can be used by autonomous robots on Mars to search for rocks that contain carbonates and which therefore might indicate a geological environment that may once have supported life. In order for such autonomous science to be possible, one needs to have a reliable detector of carbonates. The goal of Rockspec is to help determine which algorithms can make the best use of information contained in spectral data to do classification of this sort.