Electromyographic signal classification of writing selected Latin letters and numbers by artificial neural network

In the world of medical science and technology, the analysis of electromyography (EMG) signals is pivotal for the advancement and development of smart technologies. EMG is also employed as an assessment tool in applied research areas such as physiotherapy, rehabilitation, direct muscle observation, muscle fatigue evaluation, and the examination of abnormal muscle disorder patterns. Handwriting recognition through surface EMG data is a topic that can have extensive applications in technology and authentication systems. In this study, all participants were confirmed to be healthy without any nerve or muscle issues. We have a ۶-class classification task where classes ۱ to ۴ represent the letters A to D, and classes ۵ and ۶ represent handwritten digits ۱ and ۲. We asked to write each classification, ۳۰ times within a specified time frame in a noise-free environment. Subsequently, ۷ features were extracted from the signals and tested and trained in an artificial neural network (ANN) classifier. The results indicated an accuracy of ۸۹.۱۲% in data differentiation, demonstrating the high discriminative power of the algorithm despite the weak signals and presence of miscellaneous noise. Additionally, a comparison of data across all participants showed a variance of ±۴.۵%, indicating the high precision and repeatability of the classification system. Ultimately, the application of this method is recommended for recognizing numbers and letters in tool-free writing systems and diagnosing patients with nerve and muscle disorders by their handwriting.

کلیدواژه ها:

sEMG ، ANN ، noise ، Handwrite ، Arduino