Design Of Wireless Acquisition System Based On Mechanomyogram Signal And Research On The Initial Application Of Sign Language Recognition

Mechanomyogram Signal(MMG)refers to the low-frequency vibration produced bymuscles during contraction,and can be used as a specific manifestation of human motion.Sign language recognition plays an important role in normal communication between ordinary people and deaf-mute people,and has made a lot of progress in modern resear-ch.At present,most of the research on sign language recognition is based on the acquisition of images by cameras,and some of them are based on the acquisition of relevant signals by electromyographic sensors,bending sensors and other sensors.The method of sign language pattern recognition based on muscle tone signals has not been reported yet.A wearable wireless signal acquisition system based on MMG for sign language was designed.The ADXL355 sensor was used to collect the MMG signals of the four muscles includeing extensor digitorum(ED),flexor carpi radial(FCR),flexor carpi ulnaris(FCU)and extensor carpi radialis(ECR).The expanded minimal system of STM32F405 chip was used as the main control module,and the data was transmitted to the PC through the WiFi module of E103-W02.The PC received MMG data through the designed MATLAB Graphical User Interface(GUI).Then,based on this system,sign language recognition experiments were designed to classify and recognize 18 gestures,such as thumb extension,index finger extension,middle finger extension,pinky finger extension,five-finger clenching,five-finger extension,wrist upper extension,wrist lower extension,wrist outward extension,wrist inward extension,wrist outer rotation,wrist inner rotation,’shang’,’men’,’bu’,’ni’,’wo’,’hao’.In this paper,software pretreatment was carried out on the collected four arm action signals.Teager-Kaiser energy(TKEO)and envelope were used to segment the original data into single frame action,its time domain,frequency domain and time-frequency domain features were extracted,and different feature combination methods were designed.Quadratic Discrimination(QDA),Random Forest(RF)and Support Vector Machine(SVM)were compared for classification.The results show that TKEO segmentation,combination of time domain features and wavelet singular value features,PCA dimensionality reduction and SVM classifier are the best methods for sign language recognition.The recognition rate can reach 94.72%.The recognition time is short,and the recognition efficiency is high.The experimental results verify the validity of the wearable wireless MMG signal acquisition system in the application of gesture and sign language recognition.