In order to strengthen the gradual features of ECG signals and improve the classification accuracy of time series signals, this paper proposed an adaptive algorithm IGF+Bi-LSTM based on Bi-LSTM which integrated the features of gradual data. The algorithm adaptively selected the features of gradual data with the highest degree of similarity within a certain range. Through data fusion, it enhanced the interaction of gradual features in the network hidden space and expanded the Bi-LSTM information transmission mode. Moreover, this paper proposed an improved B-distance based on difference to solve the problem of periodic mismatch and strength inconsistency between time series signals, which not only characterized the difference degree S between the target data and different label data, but also adaptively adjusted the fusion coefficient in IGF+Bi-LSTM. The experiment shows that the classification accuracy of IGF+Bi-LSTM algorithm on ECG dataset is 98.7% and the F₁ value is 98.7%, it proves the effectiveness and practicability.

[2023-02-03] Accepted Paper /doc/2022.10.0637/v1

[2023-07-05] Printed Article /doc/2022.10.0637/v2