Research Article
Fractional Time-Frequency Scattering Convolution Network
@INPROCEEDINGS{10.1007/978-3-030-90196-7_7, author={Jiabin Zheng and Jun Shi and Gong Chen and Weiping Chen and Zhenya Geng}, title={Fractional Time-Frequency Scattering Convolution Network}, proceedings={Artificial Intelligence for Communications and Networks. Third EAI International Conference, AICON 2021, Xining, China, October 23--24, 2021, Proceedings, Part I}, proceedings_a={AICON}, year={2021}, month={11}, keywords={Time-frequency scattering Scattering network Short-time fractional fourier transform Non-stationary signal analysis Translation-variant filtering}, doi={10.1007/978-3-030-90196-7_7} }- Jiabin Zheng
Jun Shi
Gong Chen
Weiping Chen
Zhenya Geng
Year: 2021
Fractional Time-Frequency Scattering Convolution Network
AICON
Springer
DOI: 10.1007/978-3-030-90196-7_7
Abstract
The wavelet scattering convolution network (SCN) have recently developed as a kind of effective feature extractor, which has achieved a great performance in signal and image processing applications. Unfortunately, as feature extractor, SCN is not appropriate to mimic the visual system of mammals in image classification tasks, so that STFT-based time-frequency scattering convolution network (TFSCN) is proposed. However, TFSCN is limited by a major drawback: it is only available for stationary signals’analysis but not for non-stationary ones, since STFT can viewed as linear translation-invariant filters in the FT domain intrinsically. The aim of this paper is to overcome this weakness using the short-time fractional fourier transform (STFRFT) which is a bank of linear translation-variant bandpass filters and thus may be used for non-stationary signal analysis. First, We present the fractional time-frequency scattering transform based upon the STFRFT. Then a generalization of TFSCN’s structure dubbed FRTFSCN is illustrated. The significant performance of FRTFSCN are shown via experiment simulations.
