GAN-based data augmentation for the classification of remote sensing multispectral images

Abstract

Multispectral images frequently suffer from limited labeled data, which constrains the accuracy of classification. The objective of data augmentation is to improve the performance of machine learning models by artificially increasing the size of the training dataset. This paper introduces mDAGAN, a data augmentation method for the classification of high resolution multispectral remote sensing images. It is an adaptation of DAGAN (Data Augmentation GAN) to multispectral images, a generative adversarial network that consists of a generator and a discriminator. The augmentation capacity of mDAGAN for three different classical supervised classification algorithms has been evaluated over three high resolution multispectral images of vegetation, providing increased classification accuracies.