Training Data Augmentation with Images Embedded with Vector Fields

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Mathematics

Date of Award

Summer 8-18-2025

Abstract

One critical stage in machine learning is the training phase, where the model learns to perform a specific task based on training data. The quality and quantity of training data significantly affect the model performance. However, in some situations, training data may be insufficient. To remedy such a situation, data augmentation approaches use existing small dataset to generate artificial data samples which process similar properties as the original samples. In this way, by uniting the two sets, the original and the generated, an augmented training data set is created. It improves the classification statistics of the machine learning classifier. In our approach, we generate new sets using the original one, where we embed vector fields (VFs). For this purpose, we apply an existing VF generator to embed VFs into images. We validate the new data augmentation approach using three public image databases, COIL 100, Fashion MNIST and Digit MNIST, and classify them using a Convolutional neural network (CNN) model. The results demonstrate that the augmented training databases outperform the classification trained with the original data set only.

Advisor

Nikolay Sirakov

Subject Categories

Mathematics | Physical Sciences and Mathematics

Share

COinS