Research On Novel Feature Recognition Technique For Complex Motor Imaginary Potential Of Lower Limbs Action

Complex motor imaginary potential (CMIP) of lower limbs action is the endogenous EEG induced by the preparation or planning of complex movements of lower limbs action. It can reflect the movement-related subjective consciousness of mutual integration and synergies between different functional areas of the brain. CMIP of lower limbs action can be used as the"window"for deep exploration on complex cognitive function, such as walking or switching between sitting and standing. It also has the irreplaceable importance in the fields of brain-computer interface, rehabilitation engineering and etc. CMIP of lower limbs action is related with more sensory-motor cortex areas and involved with more complicated patterns. Therefore, not only higher spatial resolution, but also more appropriate characteristic description method was required for better reflecting the actual characteristics of CMIP of lower limbs action.Independent component analysis (ICA) is commonly used for improving the EEG spatial resolution, but it has limitations on assuming the spreading process of EEG a simple linear and instantaneous mixing one and omiting the spatio-temporal dynamics of underlying neural processes, this leads to the constraint for further improving the spatial resolution. As for characteristic description methods, empirical mode decomposition (EMD) method was put forward to reflect the signal components by the self-adaptive basis function according to the signal itself firstly. A new concept of signal components is defined based on the oscillation modes of the signal which brought a new revolution in signal processing area. But EMD's limitations on model mixture requires its further improvements.In this study, the CIMP induced by 3 key complex imaginary movements of lower limbs action (imaginary standing up, imaginary collaboration action of left hand and left leg, imaginary collaboration action of right hand and right leg) were analyzed based on the convolved mixed model, then the spatial filtering method of wavelet packet-based independent component analysis was established to meet the spatial resolution requirements for CMIP of lower limbs action; An improved EMD combined with multi-resolution analysis was designed for feature extraction of CMIP, and the feature scales were classified to avoid the model mixture; The characteristics of the signal components was defined through the basis functions constructed by the characteristics of CMIP of lower limbs action under alpha and beta rhythm, this provides a new approach to extract ERD/ERS features from the view of feature oscillations; The relationship on integration and collaboration between different sensori-motor regions was studied by phase synchronization features of characteristics oscillation mode; Finally, CMIPs of lower limb actions were identified and classified by the multi-classification algorithm based on support vector machines and characteristics combination algorithm.