This document proposes using generative adversarial networks (GANs) to generate additional training data samples for a convolutional neural network (CNN) classifier in order to address problems with limited training data and conditions not covered in the original dataset. Specifically, the author proposes training a GAN model to generate new image samples under different conditions and combining this synthetic data with the original dataset to train the CNN. The accuracy of the CNN will then be compared when trained on just the original data versus when trained on the combined original and synthetic datasets. This approach aims to improve the CNN's performance by expanding the variety of training conditions it sees through data augmentation with GANs.