Improved method For Generative Adversarial Nets

Abstract

Recently, deep learning has developed rapidly and contributed in many fields like the classification in radar and sonar applications. In some special fields like the underwater acoustic signals, the dataset for training may be scarce due to the reason of security or other restrictions, which affects the performance of the deep learning methods as those need a big dataset to ensure high accuracy. Furthermore, the original dataset is in some formats like audio, which makes those methods difficult to capture features, especially in insufficient sample case because of the interference. This paper presents a novel framework that applies the LOFAR spectrum for preprocessing to retain key features and utilises improved Generative Adversarial Networks (GANs) for the expansion of samples to improve the performance classification. Traditional convolutional GANs generate high-resolution details as a function of only spatially local points in lower-resolution feature maps. In our method, details can be generated using cues from all feature locations. Moreover, the discriminator can check that highly detailed features in distant portions of the image are consistent with each other. The experimental results show that the generated samples have high quality, which can significantly improve the classification accuracy of the neural models.