基于动作特异图卷积与注意力机制的行为识别方法

打开文本图片集

中图分类号：TP242 文献标志码：A 文章编号： 1000-5013（2026）01-0083-10

Abstract：To address the challenges human action recognition on resource-constrained edge devices，enhance model robustness， allviate confusion among similar actions，a behavior recognition network based on action-specific graph convolution attention mechanism network （ASGCA-Net） is proposed. In the temporal dimension，a multi-scale temporal convolutional attention module is designed to simultaneously capture short-term local motion paterns long-term global dependencies，thereby strengthening the temporal modeling capability action sequences.The attention mechanism is further employed to learn the weight importance each channe. In the spatial dimension，implicit edges are introduced into the fixed topology to supplement joint dependencies across physical connections， a gating mechanism is used to adaptively adjust the weights structural implicit edges，enabling action-specific feature modeling for different action types. Finally，ASGCA-Net is evaluated on the NTU RGB+D the NTU RGB+D 120 datasets. The results show that compared with the baseline network 2s-AGCN，ASGCA-Net achieves substantial accuracy improvements on both datasets.

words：human action recognition；graph convolution； temporal convolution； attention mechanism

行为识别是计算机视觉与模式识别的重要研究方向，旨在让机器理解和分析人类的动作模式。（剩余16112字）