Autonomous rock classification using Bayesian image analysis for Rover-based planetary exploration. (October 2015)

• Record Type: Journal Article
• Authors: Sharif, Helia; Ralchenko, Maxim; Samson, Claire; Ellery, Alex
• Is Part Of: Computers & geosciences. Volume 83(2015)
• Page Start: 153
• Page End: 167
• Publication Date: 2015-10
• Subjects: Autonomous geology -- Textural analysis -- Bayesian network -- Haralick parameter -- Planetary exploration -- Exploration Rover; Environmental policy -- Periodicals; 550.5
• DOI: 10.1016/j.cageo.2015.05.011
• Languages: English
• ISSNs: 0098-3004

Abstract

Abstract: A robust classification system is proposed to support autonomous geological mapping of rocky outcrops using grayscale digital images acquired by a planetary exploration rover. The classifier uses 13 Haralick textural parameters to describe the surface of rock samples, automatically catalogues this information into a 5-bin data structure, computes Bayesian probabilities, and outputs an identification. The system has been demonstrated using a library of 30 digital images of igneous, sedimentary and metamorphic rocks. The images are 3.5×3.5 cm 2 in size and composed of 256×256 pixels with 256 grayscale levels. They are first converted to gray level co-occurrence matrices which quantify the number of times adjacent pixels of similar intensity are present. The Haralick parameters are computed from these matrices. When all 13 parameters are used, classification accuracy, defined using an empirical scoring system, is 65% due to a large number of false positives. When the number of parameters and the choice of parameter is optimized, classification accuracy increases to 80%. The best results were achieved with 3 parameters that can be interpreted visually (angular second moment, contrast, correlation) together with two statistical parameters (sum of squares variance and difference variance) and a parameter derived from information theory (information measure of correlation II). The system has been kept simple not to draw excessive computational power from the rover. It could, however, be easily extended to handle additional parameters such as images acquired at different wavelengths. Highlights: Developed a robust visual rock classification system for planetary exploration. Textural information is extracted from images via 13 Haralick parameters. Bayesian probabilistic theorem best matches the samples with pre-catalogued data. Using an optimized subset of 6 Haralick parameters, the system offers 80.3% accuracy. System could provide autonomous science capabilities to a planetary rover.