T2I-Net: Time Series Classification via Deep Sequence-to-image Transformation Networks

One-dimensional convolutional neural networks have achieved great success in time series classification problems due to their ability to automatically learn feature representations. Therefore, the mainstream researches focus on the optimization of 1D neural network structure to enhance the time series classification accuracy. In this paper, we propose a novel structure considering 1D time series and 2D images transformed by 1D time series for classification. Furthermore, from a computer vision view, the influence on the accuracy of time series image transformed methods and the image sizes are discussed. The experimental results on time series classification datasets show that our proposed model has competitive results, compared to the 1D convolutional approach and 2D convolutional approach on time series classification.