Feature representation is a crucial part of time series data mining, which directly affects the accuracy of series data classification. Time series data often have average results using traditional methods for pattern classification because of the complexity of the modes and the features that can be used for classification are not easily extracted. For this reason, based on image processing techniques, a method is proposed to graph time series and combine them with image moment features for classification. The method converts time series data into binary images by gridding them, then extracts image moment features and imports them into a classifier for pattern classification, and finally achieves similarity assessment of the series data. The experimental classification accuracy of the method on the UCR dataset outperforms other algorithms and proves the effectiveness of the method.
