Pressure Pain Recognition for Lower Limb Exoskeleton Robot with Physiological Signals

Pain is a feedback mechanism the body uses to protect itself. The perception of human pain is still missing in lower limb exoskeleton robots, such as the pain caused by the pressure between the knee joint and the baffle when the lower limb exoskeleton stands up, or the pain caused by the mismatch between the gait and the body function of the wearer when walking. Aiming at the pain perception problem of the exoskeleton robot, this study designed a pressure pain data acquisition paradigm combined first considering the movement mode of the exoskeleton robot, and collected facial electromyography and electrocardiography signals, as well as five levels of pain. Then a deep learning neural Network based on Temporal Convolutional Network and Long Short-Term Memory is designed. The composite Loss of Center Loss, InfoNEC Loss and Softmax Loss is designed. The designed deep neural network model was evaluated using the thermal pain dataset BioVid, and the average recognition rate using only electrocardiography signals reached 86.44%. By Using the fusion signals of electrocardiography, electrocardiogram and electromyogram signals, the average recognition accuracy can reach 87.16%, which is superior to other similar methods, and verifies the effectiveness of the proposed method. The proposed network is also used to verify the collected pressure pain dataset, and the recognition accuracy can reach 87.5%, which proves the validity of the collected data set.