Convolutional Neural Networks (CNN) has achieved a great success in imagerecognition task by automatically learning a hierarchical featurerepresentation from raw data. While the majority of Time-Series Classification(TSC) literature is focused on 1D signals, this paper uses Recurrence Plots(RP) to transform time-series into 2D texture images and then take advantage ofthe deep CNN classifier. Image representation of time-series introducesdifferent feature types that are not available for 1D signals, and thereforeTSC can be treated as texture image recognition task. CNN model also allowslearning different levels of representations together with a classifier,jointly and automatically. Therefore, using RP and CNN in a unified frameworkis expected to boost the recognition rate of TSC. Experimental results on theUCR time-series classification archive demonstrate competitive accuracy of theproposed approach, compared not only to the existing deep architectures, butalso to the state-of-the art TSC algorithms.
展开▼
