“…In the literature [17,18], constellation map rotation, flip, and random erasing algorithms have been cited as data augmentation techniques combined with LSTM, which eventually achieved a recognition accuracy of about 92%, separately. In [19], a hybrid neural network model MCBL was proposed, which combined a CNN, bidirectional long short-term memory network (BLSTM), and attention mechanism, utilizing their respective capabilities to extract the significant features of space and time embedded in the signal samples; the recognition accuracy reached 93%. To overcome the limitations of a small number of training samples, manual feature extraction, and low recognition accuracy, we proposed a recognition method that combined time series data augmentation and a spatiotemporal multi-channel framework.…”