Large volume and high-quality training datasets are fundamental of high performance data-driven techniques. Since it is computationally expensive to obtaining data through physical experiments, instruments, and simulations, data augmentation techniques for scientific applications are becoming a new direction to obtain additional scientific data recently. However, existing data augmentation techniques from computer vision, yield unrealistic data samples that are not helpful for the domain problems that we are interested in. In this paper, we develop new physics-informed data augmentation techniques based on convolutional neural networks. Specifically, our methods leverage observable perception to improve the quality of the synthetic data.