基于密集图卷积和注意力的攀爬行为识别技术  

Climbing behavior recognition technology based on dense graph convolution and attention

作　　者：姚砺[1] 魏钰菁 万燕[1] YAO Li;WEI Yujing;WAN Yan(College of Computer Science and Technology,Donghua University,Shanghai 201620,China)

机构地区：[1]东华大学计算机科学与技术学院,上海201620

出　　处：《智能计算机与应用》2024年第6期50-55,共6页Intelligent Computer and Applications

摘　　要：为解决大量特征信息在经过特征提取网络层层传递过程中不断被削弱以及针对时空和通道信息提取不充分的问题,本文基于MST-GCN提出了基于注意力和密集图卷积的攀爬行为识别方法。首先,在MST-GCN原先的时空图卷积网络中引入了密集连接,重构了特征提取网络,获取更丰富的关节之间的上下文关系;其次,在每层卷积块中引入卷积块注意力模块(CBAM),沿着通道和空间维度生成注意力特征图,加强模型对关键信息的特征提取能力。实验结果表明,本文所提出的方法相对于基线网络对攀爬行为的识别准确率提升了11.2%,并且超过当前其他方法。To address the problem that a large amount of feature information is continuously weakened during the process of feature extraction network layer-by-layer propagation,as well as the insufficient extraction of spatiotemporal and channel information,this paper proposes an attention-based and dense Graph Convolutional Network(GCN)method for rock climbing behavior recognition based on the MST-GCN.Firstly,dense connections are introduced into the spatial-temporal GCN network of the MST-GCN,reconstructing the feature extraction network to obtain a more comprehensive contextual relationship between joints.Then,the Convolutional Block Attention Module(CBAM)is introduced into each layer of the convolutional block to generate attention feature maps along channel and spatial dimensions,enhancing the model's ability to extract key feature information.Experimental results show that the proposed method in this paper improves the recognition accuracy of climbing behavior by 11.2%compared with the baseline network,and surpassing other current methods.

关 键 词：骨架攀爬行为识别 密集连接 CBAM 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]