Research On EEG Emotion Recognition Technology Based On FPGA

The research of emotion recognition methods based on EEG signals can contribute to accurately assess the emotional state of patients with autism,depression,and other conditions.In recent years,the demand for applying emotion recognition methods to hardware devices increases with each passing day,and FPGA has become a research hotspot for hardware implementation of many methods due to its advantages such as low power consumption,reconfigurability,logic simplification,and high parallelism.Therefore,the research of EEG emotion recognition method based on FPGA is carries out,the main research work includes:(1)A multi-channel parallel network based on discrete parallel processing loop mode is proposed to address the problems of channel redundancy and high computational overhead of traditional deep convolutional neural networks.This network includes a multi-channel discrete parallel processing module,a multi-channel information exchange and reorganization module,and a weighted classification module.Firstly,different channels of the input images are processed in parallel based on multi-channel discrete parallel processing module to capture the specific feature information of each channel.Secondly,the feature maps of different channels are exchanged and reorganized based on multi-channel information exchange and reorganization module to achieve information exchange across groups,thus extracting combined features with more characterization ability.Finally,feature aggregation and classification are achieved based on the weighted classification module,and a channel weighted pooling layer is used instead of the fully connected layer in this module to reduce abundant parameters and achieve further filtering of fused features.(2)A novel emotion recognition method based on multi-band EEG topology representation is proposed aiming at the problem that most emotion recognition algorithms do not consider combining the frequency features of EEG signals with spatial information and frequency band characteristics.In this method,the multi-band EEG topology maps are used to present the state changes of EEG signals in the frequency domain,spatial domain and frequency bands to provide richer information related to the emotional states,thus improving the recognition performance.(3)An emotion recognition accelerator based on FPGA is proposed.In this accelerator,the convolutional operation module,data processing module,and dense computing module are combined with each other based on the instruction information sent by ARM to complete the entire emotion recognition task.Aiming at the problem of how to bring the parallel computing ability in convolution operation into full play,several multi-kernels processing elements are used to realize parallel operation of different channels of feature map,and the parallelism within sliding window and of the output channel of the convolutional operation is realized by the loop unrolling technology to realize the hardware acceleration of the convolutional operation.