AC-WGAN-GP: Generating Labeled Samples for Improving Hyperspectral Image Classification with Small-Samples

The lack of labeled samples severely restricts the classification performance deep learning on hyperspectral image classification. To solve this problem, Generative Adversarial Networks (GAN) are usually used for data augmentation. However, GAN have several problems with task, such as poor quality generated and an unstable training process. Thereby, knowing how to construct a generate high-quality is meaningful small-sample task data. In paper, Auxiliary Classifier based Wasserstein Gradient Penalty (AC-WGAN-GP) was proposed. framework includes AC-WGAN-GP, online generation mechanism, sample selection algorithm. proposed method has following distinctive advantages. First, input generator guided by prior knowledge separate classifier introduced architecture AC-WGAN-GP produce reliable labels. Second, mechanism ensures diversity samples. Third, that similar real selected. Experiments three public datasets show follow same distribution enough diversity, which effectively expands set. Compared other competitive methods, achieved better accuracy small number