Mineral Information Extraction Using Reflectance Spectra Of Rocks And Modified PC Algorithm

So far as planetary exploration is concerned, visible and near infrared reflectance spectra data(from approximately 0.4μm to 2.5μm) in particular has offered geologists an important potential source of petrological information.However,visible and near infrared spectra of rocks or minerals are susceptible to be affected by environmental or uncertain factors.As a result, the stability and correctness of many traditional spectral discriminate methods are not guaranted.At present,Bayesian models are becoming increasingly prominent across a broad spectrum of the cognitive sciences. Just in the last few years, Bayesian Networks have addressed animal learning, causal learning and inference. However, we have found few published systematic(or unsystematic) study exploring its reliability on the task of practical mineral identification from visible to near infrared reflectance spectra.So we use a Bayesian network algorithm which we refer to as modified PC algorithm for constructing causal Bayes networks and detecting rocks’ spectra.In our test,we use field spectra of rocks in Lamasu Copper Deposit as a test set and the USGS library as a reference set.However,the modified PC algorithm is proved to be useful for identifying spectra of various types of rocks through our experiment.The modified PC procedure, in combination with a data filter restricting the set of wavelengths, performs better than the algorithm in combination with no data filter. Raw spectral data is preprocessed in our analysis in a variety of ways so that it can usefully be analyzed by midified PC algorithm. In addition, we compare the reliability of the modified PC with that of Spectral Angle Mapper method and Spectral Feature Fitting method.In the detection of field rocks’ data, the three algorithms show different advantages, so we can use them together to improve the accuracy of detection.On the other hand,the modified PC algorithm performs more reliable than Spectral Angle Mapper method and Spectral Feature Fitting method in the detection of reflectance spectra of 135 large-grain mineral samples in the JPLlibrary.Finally, we put some advice to improve correctness and practicability of modified PC algorithm.