Lithological composition sensor based on digital image featureextraction, genetic selection of features and neural classification

摘要

A computer vision system is under development to classify thelithology of rock material on a conveyor belt in a mineral processingplant. The objective of the system is to classify the lithology of thematerial by considering seven common lithological classes found in theore: turmaline breccia, other breccias, porphyritic dykes, daciticdiatreme, granodiorites, andesite and riolitic diatreme. The informationabout the ore lithological composition will help optimize the grindingactivity of the plant. A database of 760 digital images of the sevenlithological classes was developed. A segmentation procedure wasdeveloped to isolate individual rocks. A set of 130 features wasextracted from each segmented rock of the database. The geneticalgorithm selected 70 of the 130 extracted features with no significantloss in classification performance measured in the test data set. Thereduction of the number of inputs also reduced the computation time forfeature extraction by nearly 50%