Data Augmentation to Improve the Performance of a Convolutional Neural Network on Image Classification
Deshan Fonseka, Christos Chrysoulas · 2020 International Conference on Decision Aid Sciences and Application (DASA) · 2020
Deep learning has become a fundamental tool to extract meaningful information from big data. However, it needs a huge amount of high-quality data to build an accurate classifier. In many situations, the size of the training dataset is not sufficiently large to effectively train a model. This paper presents a Convolutional Neural Network trained on a very small dataset, discussing the impact of data augmentation, feature extraction and fine-tuning on the accuracy of the model. The results show that having a small dataset, those approaches are very effective when dealing with image data.