Hybrid Attention-Based Shifted Graph Convolutional Networks
Bo Liu, Guoping Wang, Zhang Aihua, Xiaoqun Liu, Zhiyin Han · Research Square · 2023
Abstract [1]Aiming at the problem that the existing algorithms for recognizing human actions in videos will be affected by changes in the environment or human body posture, a mixed-domain attention channel is designed using channel attention and spatial attention mechanisms to optimize the feature data input by the network. Use the mixed attention mechanism to detect specific parts and increase the proportion in the skeleton map, thereby improving the accuracy of action recognition. Secondly, using the shift operation to replace the conventional convolution operation in the space-time convolution graph can obtain a convolution effect that is more efficient, has fewer parameters, and has a larger receptive field. This paper conducts experiments on the NTU RGB-D dataset, and the recognition accuracy of human actions on the X-Sub subset reaches 94.5%. [1]This work was supported by the Natural Science Foundation of Hebei Province (hky22027). AuthorResume: Liu Xiaoqun (1968-), male, Zhangjiakou, Hebei Province, professor, master director, main research interests: computer network and information security.