Slope recognition based on human body surface EMG signal Using CNN

In recent years, the development of intelligent exoskeleton robot technology has made considerable applications in the military and civilian fields. Accurate recognition of human motion patterns and compliant switching of control systems are technical difficulties to be solved in the field of intelligent exoskeleton. Convolutional neural networks (CNN) have achieved good applications in the fields of computer vision and speech recognition. Practice has proved that slope detection is an important part of human motion pattern recognition. However, few people are engaged in related research. In this paper, in view of the fact that the surface EMG signal is generated before the action and is similar to the audio signal, we introduce a slope-recognition method based on the raw surface EMG signal using CNN. Without using the other feature extraction and signal processing methods, we use short-time Fourier transform (STFT) to process the original EMG signal to generate a spectrogram as CNN input. As a result, compared with traditional machine learning algorithms, our method has a higher accuracy of 99.94%, which is vital for exoskeleton robots that directly interact with the human body due to safety and comfort.