Research On Motor Imagery Classification Algorithm Based On Deep Transfer Learning

The brain-computer interface is a system that can directly measure the brain activity related to the user's intention,convert it into a control signal,and then control an external device.Motor imagery,one of the paradigms of brain-computer interface,is a way of thinking that imitates the motor intention without real motion output,that is,the brain imagines the entire movement without contracting the muscles.Research has shown that motor imagery can produce the same change of sensory motor rhythms as a real movement.In the motor imagery based brain-computer interface system,the stroke patients image specific actions to control external devices to finish the rehabilitation training.Motor imagery can continuously stimulate the damaged motor cortex to reactivate the motor nerve cells around the damaged cells and partially restore the patients' motor function.At present,deep learning algorithms have received extensive attention in areas such as feature extraction and pattern recognition in brain-computer interface systems.Because the process of EEG signal collection is prone to problems such as movement,lack of concentration and the cost of data labeling is higher,it affects the construction of large-scale data sets.In the motor imagery based brain-computer interface system,the application of deep learning algorithms is limited.In the paper,used a deep transfer learning method to solve this problem.The purpose of transfer learning is to apply knowledge or patterns learned from one task to other different but related tasks.The specific content of this paper is as follows:The paper proposed an end-to-end transfer learning method based on the EEGNet model to identify different motor imagery tasks.Explored deep transfer learning in two types of data intra-subject and inter-subject.Used multiple convolutional neural networks to decode the electroencephalogram EEG of stroke patients and healthy subject.Introduced the fine-tune in transfer learning to improve the model,and used the improved model in the feature extraction and pattern recognition of EEG signals.Discussed the complexity of each neural network model,the complexity of the algorithm is reduced and the classification result is also improved,which verifies the effectiveness of the algorithm.At the same time,the paper proposes a transfer learning method based on the VGG16 model to identify different motor imagery tasks.Used Ensemble Empirical Mode Decomposition to denoise EEG signals and the wavelet transform changed the three-channel EEG signal into a gray-scale time-frequency feature map,which are used as the input of the pre-training model.Introduce transfer learning technology to evaluate model classification performance,and the algorithm finally obtains a better classification result.Compared the classification results between the proposed framework and traditional classifiers such as Support Vector Machine,Gradient Boosting and the original network VGG16 model,and the effect is the best.The paper designs a paradigm of EEG acquisition based on stroke patients.Familiar with the use of EEG acquisition equipment,design a set of reasonable experimental acquisition schemes,accumulate experimental data,and lay a solid foundation for the analysis and processing of EEG signals of stroke patients.