Human poses recognition based on Spiking Pulse Graph Neural Networks

The temporal dilated convolutional model usually requires a large amount of computing resources in the human pose recognition task, especially when the input image information is too complex, which will lead to the shortcomings of low accuracy and large model energy consumption. Addressing these issues, we designed a new human poses recognition Spiking Pulse Graph Neural Networks model. In this model, we increase the receptive field of the model by changing the expansion coefficient and activation function of the convolution in the model receptive field processing module, so as to make the extracted features more accurate. The transmission rate of feature information is controlled by the Spiking Pulse Graph Neural Networks model with the reward mechanism of human pose learning rate, which is used to improve the accuracy of the model and reduce the energy loss of the model. Compared with the temporal dilated convolutional model and the latest human pose recognition method, the accuracy of human pose recognition is improved and the energy consumption is reduced under the same test set data.