RTFN: A robust temporal feature network for time series classification

Time series data usually contains local and global patterns. Most of the existing feature networks focus on features rather than relationships among them. The latter is also essential, yet more difficult to explore because it challenging obtain sufficient representations using a network. To this end, we propose novel robust temporal network (RTFN) for extraction in time classification, containing (TFN) long short-term memory (LSTM)-based attention (LSTMaN). TFN residual structure with multiple convolutional layers, functions as local-feature mine from data. LSTMaN composed two identical where LSTM are hybridized. This acts relation discover intrinsic extracted different positions. In experiments, embed RTFN into supervised unsupervised structures extractor encoder, respectively. results show that RTFN-based achieve excellent performances large number UCR2018 UEA2018 datasets.