RSMDA: Random Slices Mixing Data Augmentation

Advanced data augmentation techniques have demonstrated great success in deep learning algorithms. Among these techniques, single-image-based (SIBDA), which a single image’s regions are randomly erased different ways, has shown promising results. However, erasing image SIBDA can cause loss of the key discriminating features, consequently misleading neural networks and lowering their performance. To alleviate this issue, paper, we propose random slices mixing (RSMDA) technique, one placed onto another to create third that enriches diversity data. RSMDA also mixes labels original images an augmented label for new exploit smoothing. Furthermore, investigate three strategies RSMDA: (i) vertical strategy, (ii) horizontal (iii) mix both strategies. Of strategies, slice strategy shows best validate proposed perform several experiments using across four datasets: fashion-MNIST, CIFAR10, CIFAR100, STL10. The experimental results classification with showed better accuracy robustness than state-of-the-art (SOTA) multi-image-based methods. Finally, class activation maps employed visualize focus network compare SOTA