7 Undergraduate Research I (Posters)

Wednesday, 12 May 2021
11:25 AM - 12:30 PM
Mackay School of Earth Sciences & Engineering Room

AUTOMATIC CLASSIFICATION OF PLUTONIC ROCKS WITH MACHINE LEARNING APPLIED TO DOMINANT COLORS ON IOS DEVICES

Paper number: 7-11
Program: Theme Sessions
Time: 12:15 PM – 12:20 PM

Authors

Abstract

Plutonic rocks are formed when magma cools and solidifies below the Earth’s surface. Lightness and color are properties used for the classification of plutonic rocks; however, these attributes can be difficult to describe because perceived rock colors depend on the observer’s experience. Moreover, although the classification of plutonic rocks can be done using data from various instrumental techniques, these approaches tend to be expensive and time-consuming.

This research extracts dominant shades and colors from plutonic rock images to train several machine learning algorithms and deploy the best model in an iOS app for the automatic classification of four classes of plutonic rocks in order from darker to lighter: gabbro, diorite, granodiorite, and granite. We used pictures from plutonic rocks that were classified by using petrography and chemistry data.

First, the dominant colors of plutonic rock images were extracted with the k-means algorithm by grouping the image pixels according to the RGB and CIELAB color spaces. Then, the data of the four dominant colors were used to create and evaluate several machine learning models with the following algorithms: Logistic Regression, K-Nearest Neighbors (KNN), Decision Trees, Support Vector Machine, and Convolutional Neural Networks. The experiments were executed first with the dominant colors in RGB and then in CIELAB. The best model was deployed after validation on an iOS application that classifies the extracted colors in new images of the four rock types.

The best results during validation were for the model generated with KNN trained with 283 images in the CIELAB color space. Results gave accuracy, precision, recall, and F-score average values of 93%. The application running the KNN model was tested in the field with 34 images, and the following average accuracy results were obtained: 70% for gabbro, 28.5% for diorite, 20% for granodiorite, and 85.7% for granite. The high accuracy when classifying gabbro samples was because they are noticeable darker than samples of the other 3 classes. Similarly, granites were noticeably lighter. In contrast, diorite and granodiorite share characteristics of the other rock types closest to them in the dark-light sequence; therefore, it is more difficult to automatically classify them based on their dominant colors.

Geological Society of America Abstracts with Programs. Vol. 53, No. 4

doi: 10.1130/abs/2021CD-362983

© Copyright 2021 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.