This work investigates digital image processing techniques for the automated identification of minerals in rock thin sections, i.e., in the microscopic scale. More specifically, algorithms have been developed that identify minerals and rocks in thin sections with the use of polarized light. A large number of thin sections of different types of rocks (plutonic, metamorphic, etc.) have been captured under the petrographic microscope in circularly polarized light conditions, in order to avoid mineral extinction. These images are taken under specific camera parameters (Correlated Color Temperature and Exposure Time), with defined luminance and stored in a database. Textural, albedo, and color features are extracted, including the average, standard deviation, roughness, skewness, contrast, etc. Furthermore, deep learning techniques (Neural Networks, SVM, KNN, etc.) have been developed and tested in order to choose the most efficient method. Following of the mineralogy identification, the rocks are then classified according to their type. These computer algorithms can be used in geosciences in order to detect micro-textural features such as the degree of crystallization. Our ultimate goal is to integrate such a system on a rover for autonomous exploration or even for the petrographic study of rocks when the in situ production of rock thin sections will be possible during space missions.

References

Baykan, N., A., Yilmaz, N. (2010): Mineral identificatiob using color spaces and artificial neural networks, Computers and Geociences, vol. 6, pp. 91-97

Ross, B., J., Fueten, F., Yashkir, D., Y. (2001): Automatic mineral identification using generic programming, Machin Vision and Applications, vol. 13, pp. 61-69

Aboudan, A., Pacifici, A., Murana, A., Allemand, P., Ori1, G. G., Marcer, A. Portigliotti, S., Lorenzoni, L. (2014): AUTOMATIC ROCKS DETECTION AND CLASSIFICATION ON HIGH RESOLUTION IMAGES OF PLANETARY SURFACES, in 45th Lunar and Planetary Science Conference

Dunlop, H. (2006): Automatic rock detection and classification in natural images, Master Thesis in Robotics Institute, Carnegie Mellon University, Pittsburgh, Pennsylvania

Partio, M., Cramariuc, B., Gabbouj, M., Visa, A. (2002): ROCK TEXTURE RETRIEVAL USING GRAY LEVEL CO-OCCURRENCE MATRIX, in Proc. of the 5th Nordic Signal Processing Symposium, Norway

Song, Y. (2008): AUTOMATED ROCK SEGMENTATION FOR MARS EXPLORATION ROVER IMAGERY, 39th Lunar and Planetary Science Conference, No. 1391, p.2043

Foucher, F., et. al. (2018): Lithospace: an automated system for in situ petrographic thin section preparation on Mars, EPSC Abstracts vol. 12, EPSC2018-1172