| 摘要: | 本計畫將研製一套結合雲端穿戴訓練裝置之腦波人機介面行動輔具，提供行動不便的病患一個外 骨骼裝置的輔助器材，藉由判斷使用者想像運動腦波的變化，與受測者配戴的重力感測器(G-sensor) 實際運動時的訊號做事前的比對，並利用全息希爾伯特譜分析法(Holo-Hilbert Spectral Analysis, HHSA) 演算法分析腦波訊號的成分，分離其主要頻率與振幅調變(Amplitude Modulation, AM)之成分，透過演 算法的計算得到相對應的馬達控制訊號，以控制行動輔具。 本計畫開發之腦波控制機械外骨骼系統主要分成兩個階段，第一階段為腦波與動作狀態的訓練， 透過無線傳輸的方式將重力感測器的數值傳回至主要運算單元，傳輸的部分需考慮多通道的訊號間干 擾問題與空間中無縫傳輸。得到數值後，利用卷積神經網路(Convolutional Neural Network, CNN)與量 測到的腦波訊號進行比對訓練，找出對應方式。第二部分則是將想像運動的腦波訊號做為系統輸入， 經即時分析後得到輔具系統的軌跡命令，透過FPGA 同步傳給多個馬達，實現腦波人機介面行動輔具。 本計畫規劃使用可程式系統晶片(System on Chip, SoC) 結合FPGA 的系統作為控制核心，將處理器強 大的軟體功能與FPGA 靈活的硬體設計結合在一起，DSP 或MCU 負責較複雜的演算法，再利用FPGA 同步平行處理的優點，提升系統的效率。 ;This project aims to develop a Brain Computer Interface (BCI) assistive devices, EEG-controlled robotic exoskeleton system, with cloud-integrated wearable training equipment, which provides handicapped people an exoskeletal suit used in their daily life. At the beginning, we have to measure electroencephalography (EEG) signal when subject are doing imagery exercise, then compares those data with G-sensors’ signal when subject are doing real exercise. Traditionally, Hilbert-Huang Transform (HHT) is widely used for analyzing brain wave signal, however, for nonlinear processes, it can only tell the additive part. By implementing Holo-Hilbert Spectral Analysis (HHSA) algorithm, the original EEG signal can be separated into two major parts: main frequency and its amplitude modulation affected by movement or other measurement problems. This method allows developer to analyze the EEG signal precisely. Through other algorithms, the corresponding motor control signals are attained for controlling the assistive devices. The proposed project can be separated into two stages: First stage focus on data training between EEG and motions. The value of each g-sensors will transmit back to main processor through wireless way. It needs to be considered that it has interference when several channels transmit at the same time. We also aim to build seamless transmission. After having g-sensor’s value, convolutional neural network is required to train the data with EEG signal and find patterns; Second stage takes imagery exercise EEG signal as input of the system to get motors commands, which will be send to motors by FPGA. We choose SoC FPGA (System on Programmable Chip Field Programmable Gate Array) as system’s core. DSP or MCU part will in charge of some complicated algorithms. Combined with FPGA which has advantage of parallel processing, the efficiency of the system is able to increase. |
