Abstract: Many computer vision tasks have shown that deep convolutional neural networks perform exceptionally well. However, in order to avoid overfitting, these networks rely extensively on huge amounts of data. Many disciplines, such as medical image analysis, lack access to massive data sets. Data augmentation approaches enable applications with limited datasets to achieve higher accuracy. The process of creating samples by transforming training data is known as data augmentation, with the goal of increasing classifier accuracy and resilience. Geometric transformations, colour space augmentations, kernel filters, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, and meta-learning are some of the image augmentation technologies explored in this review. Data Augmentation will help readers learn how to improve the performance of their models and expand their limited datasets in order to take advantage of big data.

Keywords: Machine learning, Deep learning, Data augmentation, GAN, Medical imaging.

PDF | DOI: 10.17148/IJARCCE.2022.11390

Open chat
Chat with IJARCCE