Convolution Neural Network (CNN)-based Upper Limb Action Recognition

Surface electromyography (sEMG) can directly reflect human neuromuscular activity, so sEMG is used to track and identify human joint movement in the field of rehabilitation medical engineering. For a large number of sEMG graphs, the results of using traditional classifiers to process the results are unsatisfactory. Using large classic convolution neural network (CNN) to process for a long time will cause a large delay in the control process. In order to make up for the shortcomings and deficiencies, the paper uses lightweight CNN to effectively classify and predict a large number of sEMGs in different parts. Since a lightweight network is used, low latency can be achieved for control effects. Theoretically, a training model with strong real-time performance and high accuracy can be achieved, and it also has a considerable effect in the control process.