Research On Upper Limb Rehabilitation Path Planning And Action Classification Based On Task-oriented Training

In recent years,the number of patients with central nervous system damage caused by cerebrovascular diseases such as stroke resulting in hemiplegia and motor dysfunction has been increasing annually,and they are becoming younger.After central nervous system injury,nerve tissue regeneration needs to be compensated by other higher central nervous systems for recovery.Based on the neuroplasticity of the brain,upper limb rehabilitation robots can perform specific motor function training to help patients achieve limb rehabilitation.However,the existing upper limb rehabilitation training programs still have shortcomings in personalized customization and practicality.To better help patients recover normal motor function,this paper focuses on three aspects: the human motion feature output function,optimization of the upper limb rehabilitation robot path,and classification of upper limb rehabilitation movements.The main research work is as follows:(1)To address the problem of task-oriented training in daily life,the human upper limb skeletal and muscle structure are analyzed from an anatomical and biomechanical perspective.Human upper limb rehabilitation motion data is collected through an optical motion capture system,and the preprocessed data is used to generate upper limb motion characteristics for rehabilitation.These characteristics are utilized to find the inherent features of the human motion framework as the basis for research on rehabilitation path planning methods aided with task-oriented training.(2)To address the problem of the optimal rehabilitation training path for the upper limb,joint activity technology is combined with the patient’s rehabilitation status at different stages to adjust rehabilitation training tasks at different stages,stimulated the recovery of different muscle groups with different training difficulties,and reshaped the central nervous system’s control of the affected limb.A comprehensive rehabilitation index based on human engineering and rehabilitation medicine is designed to constrain the human upper limb motion feature output function.A zeroing neural network is exploited to solve the time-varying nonlinear constrained optimization problem,obtaining the optimal path for upper limb rehabilitation training that can be adjusted according to the patient’s current rehabilitation stage.The task-oriented training-based optimal path for upper limb rehabilitation training can be applied to the path planning of rehabilitation robots to improve rehabilitation training efficiency,which ensure the effectiveness and comfort of upper limb rehabilitation.(3)To address the problem of cross-individual upper limb movement classification,surface electromyographic signals generated by the shoulder,elbow,and wrist during upper limb rehabilitation training are collected,and signal features are extracted to establish an action classification model based on convolutional neural networks.The upper limb rehabilitation movements are recognized and classified using this model.Experimental results show that the deep learning-based action classification model has good classification performance.To achieve the effectiveness of clinical applications,neural network pruning is used to achieve network lightening,reducing the size and space occupancy of the neural network model,and increasing the real-time performance of the upper limb rehabilitation movement classification model,and laying the foundation for clinical application of upper limb rehabilitation robots.