In recent times, deep artificial neural networks have achieved many successes in pattern recognition. Part of this success can be attributed to the reliance on big data to increase generalization. However, in the field of time series recognition, many datasets are often very small. One method of addressing this problem is through the use of data augmentation. In this paper, we survey data augmentation techniques for time series and their application to time series classification with neural networks. We propose a taxonomy and outline the four families in time series data augmentation, including transformation-based methods, pattern mixing, generative models, and decomposition methods. Furthermore, we empirically evaluate 12 time series data augmentation methods on 128 time series classification datasets with six different types of neural networks. Through the results, we are able to analyze the characteristics, advantages and disadvantages, and recommendations of each data augmentation method. This survey aims to help in the selection of time series data augmentation for neural network applications.

Image-to-image conversion tasks are more accurate and sophisticated than ever thanks to advances in deep learning. However, since typical deep learning models are trained to perform only one task, multiple trained models are required to perform each task even if they are related to each other. For example, the popular image-to-image convolutional neural network, U-Net, is normally trained for a single task. Based on U-Net, this study proposes a model that outputs variable results using only one trained model. The proposed method produces a continuously changing output by setting an external parameter. We confirm the robustness of our proposed model by evaluating it on binarization and background blurring. According to these evaluations, we confirmed that the proposed model can generate well-predicted outputs by using un-trained tuning parameters as well as the outputs by using trained tuning parameters. Furthermore, the proposed model can generate extrapolated outputs outside the learning range.