Using machine learning methods for grain composition analysis

OData support
Dr. Csorba Kristóf
Department of Automation and Applied Informatics

The subject of the thesis is the extension of the Cv4Sensorhub framework with a machine learning based classifier software component, which is capable of detecting the grain boundaries in CT images of minerals. The development of the classifier component was done by using the OpenCV library and its machine learning module, written in C#.

As my first task, I learned about the most important concepts of machine learning, and various learning algorithms, as well as the OpenCV machine learning module, gathering enough knowledge and information to successfully solve the designated problem.

In the second task, I made a general purpose information gather operator for the Cv4Sensorhub framework, to be able to export various types of information about pictures, collected along polylines.

As the third task, I designed and implemented a component for the Cv4Sensorhub, which on the one hand is suitable for hyperparameter optimization and validation of an SVM based learning model by implementing nested k-fold cross-validation, and on the other hand, is capable of classifying the pixels of mineral CT images.

As my last task, I created an appropriate dataset for my learning machine, which turned out good enough to handle the problem of imbalanced classifications.


Please sign in to download the files of this thesis.