| Abstract [eng] |
This work addresses the problem of unbalanced image data using deep neural networks. Solving this problem would lead to better results in classification tasks, which are applied in various fields such as medicine, computer vision, etc. Three main solutions to the problem of unbalanced image data at the data level are discussed - random oversampling, augmentations, and DeepSmote. Four more solutions are developed by combining the mentioned three. Experiments are carried out using different solutions and combinations of solutions, datasets, and imbalance ratios to evaluate the performance of the solutions. |
