Virtual Samples for Cloud Classification via Supervised Learning

Convolutional neural networks (CNNs) have been widely used in image classification task, which is based on the huge amount of image samples. However, the insufficiency of cloud sample numbers brings obstacles to classify clouds using CNNs. In this paper, we propose to apply Wasserstein generative adversarial network (WGAN) to generate virtual cloud samples via supervised learning. Afterward, we fine-tune a deep CNN model to evaluate the classification performance under different number of virtual cloud samples. The experimental results demonstrate the feasibility of the proposed method.