Gait recognition of hip exoskeleton for lateral walking based on ResNet

In order to meet the rehabilitation requirements of hip joint muscles, a new method of gait phase recognition based on the residual network algorithm was proposed to collect the data of electromyography and plantar pressure under lateral walking conditions. In the experiment, 10 subjects wore lateral walking exoskeletons, collected surface electromyography (EMG) and plantar pressure of 6 lower limb muscles, and extracted 8-time domain features by sliding window. Then, three kinds of convolutional residual networks of different sizes (ordinary, middle and lightweight) are constructed for classification. Among them, the middle model achieves a good balance between recognition accuracy and computational resources, while the lightweight model effectively reduces network parameters through knowledge distillation, but the recognition effect of complex categories is slightly decreased. The results show that this method can stably identify the gait phase during lateral walking, improve the training and control efficiency of the exoskeleton for hip abductor muscle, and have important reference value for lower limb rehabilitation and wearable device application.