electroencephalogram signal feature extraction and classification recognition method based on LSTM-FC

Abstract

The invention relates to a method based on LSTM-. The electroencephalogram signal feature extraction and classification recognition method for the FC is characterized by comprising the following steps: S1, collecting and preprocessing electroencephalogram signals; S2, defining a basic model structure of the LSTM network; S3, enabling the feature matrix passing through the LSTM network to pass through a two-layer FC network to obtain an LSTM model fused with the FC network, namely, LSTM-; An FC model; S4, inputting the training set into a corresponding model for training, and updating the network by using error back propagation; S5, after the corresponding model is trained, inputting the test set into the model to obtain the final classification accuracy of the motor imagery task, and thenthe performance of the model is evaluated; S6: comparing performance of LSTM and LSTM-FC models to obtain the best model.

Description

technical field [0001] The invention belongs to the technical field of brain-computer interface of artificial intelligence, and relates to a brain-computer interface (Brain- The motor imagery classification and recognition method in the Computer Interface (BCI) system, especially a method for feature extraction and classification recognition of EEG signals based on LSTM-FC; this method uses an LSTM model in deep learning to extract The characteristics of the signal and the feature matrix of the information it carries, and then use the FC network to further fuse and extract the feature information extracted by LSTM, and finally map the extracted feature matrix to the sample label space to achieve end-to-end EEG signal classification identify. Background technique [0002] BCI is a communication system that can realize the interaction between the brain and external devices. It attempts to directly connect the internal neuron activity of the brain with external devices in orde...

AI Technical Summary

Problems solved by technology

Traditional machine learning algorithms generally require cumbersome feature engineering. At the same time, the key information of time series is not considered in the modeling of complex time series signals such as EEG signals, which often leads to the loss of feature information, or the need to rely on A huge amount of calculation, and the effect of simultaneous processing is often unsatisfactory

Images