Year: 2023
Type: conference-paper
Abstract: This study concerns the classification of univariate time series. The essence of the survey is transforming time series into two-dimensional monochromatic images. Then, obtained images are classified using convolutional neural networks. Transformation of time series to images is performed in two steps. First, a time series is turned into a string of symbols from an assumed alphabet utilizing SAX-like transformation. The length of the string is supposed to be the square of a natural number. Second, the string of symbols is turned into a square matrix of size equal to the square root of the length of the string representing the time series. Then, each symbol of the matrix is turned into a square-shaped piece of pixels of a grey level determined by the symbol. So then, this operation results in an image (still of square shape) composed of squares of grey pixels. Finally, convolutional neural networks are employed to classify such images. An overall design process is presented with a focus on investigating time series-to-image two-step transformations. Experimental studies involving publicly available data sets are reported, along with an adequate comparative analyses.
DOI: 10.62036/isd.2023.56
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4390049600