Data augmentation of optical time series signals for small samples.

ABSTRACT

It is difficult to obtain a large amount of labeled data, which has become a bottleneck for the application of deep learning to analyze one-dimensional optical time series signals. In order to solve this problem, a deep convolutional generative adversarial network model suitable for augmenting optical time series signals is proposed. Based on the acoustic emission (AE) data set obtained by an optical sensor with a small amount, the model can learn the corresponding data features and apply them to generate new data. The analysis results show that our model can generate stable and diverse AE fragments in epoch 500, and there is no model collapse. All the features between the generated data and the original data are not significantly different at the 0.05 level, which confirms that the method in this paper can generate the optical time series signals effectively.