Classificação de agregados de rochas ígneas quanto a sua alteração por meio de processamento digital de imagens

Abstract

The construction of quality roads/highways depends on the choice of material that offers strength, durability and safety, described in civil engineering standards. The mineral aggregate is commonly used in large scale on paving works and, naturally, it can suffer alterations in its physical-chemical structure due to weather conditions. Such alteration in the aggregate, when used in paving works may reduce quality and durability. The objective of the present study was to investigate the use of Digital Image Processing in the classification of igneous rock aggregates, taking into account the degree of alteration, and thus allocate them in the most appropriate way for use in asphalt sidewalk construction. The materials investigated were basalt and granite. The use of texture analysis tools such as grayscale, frequency of Red, Green, Blue channels, entropy, Local binary patterns, Local binary patterns Uniform and Co-occurrence matrix have been investigated. The classifiers used were K-Nearest Neighboors, Multi-layer Perceptron, Decision Tree, Naive Bayes and Random forest. The results were submitted to statistical analysis from the Friedman and Nemenyi test to verify statistical differences. It was concluded that the texture descriptors are promising regarding to classification as to the degree of alteration of the aggregates, presenting results of 100% accuracy, in some cases, for both types of aggregates. The texture descriptors LBP, LBPU, GLCM showed good results for the group A tests, while the RGB channel frequencies and grayscale showed good performance in the group B experiments. The KNN and Random forest algorithms proved to be effective in the classification task. Promising results were also observed when granite and basalt are combined and when using samples of the aggregate or the full image.