Tactile Perception Information Recognition of Prosthetic Hand Based on DNN-LSTM

The development of prosthetic hand technology has brought good news to the patients with hand loss, but most prosthetic hands lack tactile perception, while normal human hands are very sensitive to the physical characteristics of the contact object, so the human-computer interaction is insufficient. Aiming at the problem that the membrane pressure sensor commonly used in prosthetic hand tactile measurement is difficult to obtain and distinguish a variety of tactile information, an algorithm model framework based on deep neural network (DNN)-long-term and short-term memory network (DNN-LSTM) neural network is proposed. On the basis of recognizing the grip strength of the prosthetic hand, the system model adopts the inclined shaking tactile exploration method to identify the roughness of the grasping object. First, when the prosthetic hand grabs the object normally, according to the value of the pressure sensor, the grip strength can be identified through the DNN neural network, which avoids the defect of the measurement accuracy of the traditional pressure sensor. Then, through the tilting and shaking action, according to the change of the value of the pressure sensor, the roughness of the grasping object can be recognized through the D-LSTM fusion network. Finally, through model training and real object grasping experiments, the results show that the predicted results of the trained neural network model in real grasping are in good agreement with the preset results. In summary, our proposed tactile exploration action enables a single tactile perception source to obtain other tactile information, and the proposed D-LSTM model can effectively improve the recognition accuracy.