Construção de classificador de rochas em um afloramento da Bacia Lusitânica por meio de aprendizagem de máquina

Abstract

Imagery represents a large and efficient source of information in exploration, prospecting, and geological survey. In order to obtain better and better images, the improvement of imaging techniques has become increasingly sought after and widely applied. In this, remotely piloted aircraft systems (RPAS) imagery has gained prominence due to its cost-effectiveness. Considering the economic and academic importance of correct identification and classification, to better analyze them, various machine learning algorithms are used. However, the use of machine learning requires some relevant factors, among them trained professionals in the area, computational capacity, and an extensive database about the target of interest. The main objective of this work is to create models that systematically interpret the rocks in an outcrop of the Lusitanian Basin, Portugal from images acquired with RPAS using the pixel data and create a machine learning aided classifier using the least possible number of images from the database reducing the processing time, standardizing image descriptions and analysis automating the classification of the rocks in an outcrop of the Lusitanian Basin (boundstones, rudstones, packstones conglomerates, laminated siltstones, massive siltstones and sandstones). Image segmentation models were created from the photos taken with the aerial survey with RPAS, in sequence discrete convolutional filters were used and subsequently, classifier neural networks were applied. Finally, various configurations using different models were tested. At the end of this work, satisfactory results of 88 to 90% accuracy were achieved using one third of the database.