Similar classes latent distribution modelling-based oversampling method for imbalanced image classification

Learning an unbiased classifier from imbalanced image datasets is challenging since the may be strongly biased toward majority class. To address this issue, some generative model-based oversampling methods have been proposed. However, most of these pay little attention to boundary samples, which contribute tiny learning classifier. In paper, we focus on samples and propose a similar classes latent distribution modelling-based method. Specifically, first, model each class as different von Mises–Fisher distributions, thereby aligning feature with distributions. Furthermore, develop distance minimization loss function, makes representations close other. way, generator can capture more shared features during training. addition, sampling strategy, uses variables near decision generate samples. These expand minority region reshape boundary. Experiments four show that proposed method achieves promising performance in terms Recall, Precision, F1-score, G-mean.