Unconstrained Motion Recognition Of Human Lower Limb Based On SEMG Signals

In today’s world,exoskeleton robots are commonly used.The acquisition of human motion purpose is a key aspect of human-computer interaction,which is important for the development of exoskeleton robots.The surface electromyography(sEMG)signals are generated before the movement of the limbs,which can efficiently and accurately reflect the movement intention of the human.In addition,the sEMG signals rely on the muscles associated with the movement,rather than the limbs in the executive part of the movement.It can also be collected by amputees and has a wide range of applications.Therefore,we choose to collect sEMG signals to identify human lower limb movement intention.Human lower limb motion recognition technology based on sEMG signals has been developed rapidly in various fields and received extensive attention.However,most of the existing studies focus on discrete motion recognition and ignore the importance of motion mode switching.In this thesis,human lower limb is taken as the research object,and sEMG signals are analyzed to identify human motion state and control exoskeleton robot.The main work is summarized as follows:(1)For lower limb movement intentions and complex motion pattern recognition research the problem of insufficient,we construct the unconstrained human lower limb movement dataset ULLM-sEMG,collected 18 subjects(13 males and 5 females),a total of 1440 samples,an angle-assisted annotation and recognition framework A3-SESL is proposed for this dataset,which can solve the unconstrained annotation problem of motion patterns and achieve a high recognition accuracy.(2)In order to extract the feature combinations that can represent the features of sEMG signals,we focus on the influence of different time-domain feature combinations on the pattern recognition results of sEMG signals.(3)In order to verify the validity and feasibility of the A3-SESL framework proposed in this thesis,a comparative experiment is conducted to identify the movements of unconstrained human lower limb.The experimental results show that the labeling with A3-SESL framework can effectively improve the recognition accuracy.In addition,support vector machine,multilayer perceptron,linear discriminant analysis and random forest were selected to carry out comparative experiments with A3-SESL on ULLM-SEMG dataset.The experimental results show that the recognition accuracy of A3-SESL is higher than that of other classifiers.The experiment also classifies the motion-switching modes.Further,the influence of other factors on the recognition results is explored.