Research On Lower Limb Gait And Posture Recognition Algorithm Based On Surface EMG Signals

When the human muscles are active,the brain's central system sends out control signals,which are transmitted from the axons and dendrites of spinal neurons to the end plate area.The end plate area is coupled with muscle fibers to form the basic motor unit of muscle activity.The end plate area The generated end plate potential changes the permeability of muscle cell membranes,so that many muscle fibers are spatially superimposed to generate an action potential,that is,an electromyographic signal,which is an electrical signal generated by a comprehensive reaction between the electromyographic signal and the skin surface,that is,a surface myoelectric signal.The surface EMG signal contains a lot of information about human body motion status and motion intention,which can be processed and analyzed to recognize human posture status.The human lower limb gait and posture recognition is of great significance in the fields of intelligent bionic lower limbs,medical rehabilitation,helping the elderly and the disabled.In this paper,combined with the actual movement of the human lower limbs,the surface EMG signals of the key muscles of the human lower limbs are taken as the research object,and the human gait movement patterns,surface EMG signals,experimental acquisition,signal preprocessing,and lower limb gait and posture recognition are deeply studied and discussed.Algorithms and other aspects.Specifically,this paper carried out the following research work:Firstly,the movement mode of the human lower limb gait and the generation mechanism of the surface EMG signal are described.The feasibility of the surface EMG signal for human lower limb posture recognition is analyzed.The signal data collection experiment is set up to collect different human body lower limb postures and walking Under the circumstances,the EMG signal on the thigh surface provides data support for subsequent research.Secondly,according to the noise interference in the surface EMG signal,a reasonable filtering denoising method is used,which effectively reduces the noise interference in the original signal.Combined with the characteristics of surface EMG signals,feature extraction of surface EMG signals was carried out,and a series of time-domain features were extracted from the noise-reduced signal.Through feature filtering,the autoregressive coefficients,wavelength,root mean square value,and average absolute value were determined.The combination of features is used for subsequent pattern recognition and gait recognition of the surface electromyography signals of the lower extremities.Considering that the gait data has a non-equilibrium distribution,the original sampling is used to sample the original samples to a balanced number of samples in each category.Then,the classical machine learning algorithms such as support vector machines and linear discriminant analysis,which are widely used in surface EMG signal pattern recognition,are used to classify and recognize the collected signals.Convolution is used for the characteristics of surface EMG signals that are unstable,random,and chaotic.The neural network performs non-linear feature extraction and classification recognition on the lower limb surface EMG signal.The performance of convolutional neural networks and classical machine learning algorithms on the pattern recognition task of surface EMG signals is compared and analyzed.The effect of different training sample sizes and different window lengths on the effect of each algorithm model is analyzed and discussed.Finally,classical machine learning algorithms and convolutional neural networks are used for lower limb gait recognition.The experimental results show that the one-dimensional convolutional neural network is superior to other algorithm models in terms of classification accuracy,robustness,and real-time performance.Combined with the experimental results,it is optimized for one-dimensional convolutional neural networks,using two-dimensional sliding time windows for data segmentation,adjusting the network structure,using multi-convolution kernel convolution layers of different sizes for feature extraction,and mapping the first layer of features After maximum pooling,it is spliced on the second layer feature map to improve the accuracy of network recognition,optimize the gradient descent algorithm,use the appropriate activation function and learning rate attenuation strategy to improve the network convergence speed,improve the network effect,and apply the dropout method to increase the network sparsity.To prevent overfitting,the network training speed and classification recognition accuracy have been improved.In terms of offline signal recognition,the combination of finite state machine and window filtering further reduces the error of the classification signal output by the model and obtains a more accurate recognition result.This paper systematically studies the problem of lower extremity posture and gait recognition based on surface EMG signals,and makes an in-depth discussion and analysis of the various processes that affect the problem.This paper discusses and analyzes the advantages and disadvantages of classic machine learning algorithms widely used in the field of surface EMG pattern recognition,and compares them with convolutional neural networks.Optimized for the one-dimensional convolutional neural network,and ultimately achieved an ideal improvement in terms of recognition accuracy,network convergence speed,and difficulty in setting hyperparameters.The above work is of great value to the surface EMG signal in the fields of intelligent bionic lower limbs,medical rehabilitation,helping the elderly and the disabled.