基于全局时空编码网络的猴类动物行为识别  

Monkey action recognition based on global spatiotemporal encode network

作　　者：孙峥 张素才 马喜波[1,2] SUN Zheng;ZHANG Su-cai;MA Xi-bo(CBSR&NLPR,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China;School of Artificial Intelligence,University of Chinese Academy of Sciences,Beijing 100049,China;JOINN Laboratories(Beijing)Co.,Ltd.,Beijing 100176,China)

机构地区：[1]中国科学院自动化研究所,北京100190 [2]中国科学院大学人工智能学院,北京100049 [3]北京昭衍新药研究中心股份有限公司,北京100176

出　　处：《图学学报》2022年第5期832-840,共9页Journal of Graphics

基　　金：国家自然科学基金项目(82090051,81871442);中国科学院青年创新促进会(Y201930)。

摘　　要：猴类动物行为的准确量化是临床前药物安全评价的一个基本目标。视频中猴类动物行为分析的一个重要路径是使用目标的骨架序列信息,然而现有的大部分骨架行为识别方法通常在时间和空间维度分别提取骨架序列的特征,忽略了骨架拓扑结构在时空维度的整体性。针对该问题,提出了一种基于全局时空编码网络(GSTEN)的骨架行为识别方法。该方法在时空图卷积网络(ST-GCN)的基础上,并行插入全局标志生成器(GTG)和全局时空编码器(GSTE)来提取时间和空间维度的全局特征。为了验证提出的GSTEN性能,在自建的猴类动物行为识别数据集上开展实验。实验结果表明,该网络在基本不增加模型参数量的情况下,准确率指标达到76.54%,相较于基准模型ST-GCN提升6.79%。Accurate quantification of caged monkeys’behaviors is a primary goal for the preclinical drug safety assessment.Skeleton information is important to the analysis on the behaviors of monkeys.However,most of the current skeleton-based action recognition methods usually extract the features of the skeleton sequence in the spatial and temporal dimensions,ignoring the integrity of the skeleton topology.To address this problem,we proposed a skeleton action recognition method based on the global spatiotemporal encode network(GSTEN).Based on the spatial temporal graph convolutional network(ST-GCN),the proposed method inserted global token generator(GTG)and several global spatiotemporal encoders(GSTE)in parallel to extract the global features in the spatiotemporal dimension.To verify the performance of the proposed method,we conducted experiments on a self-built monkey action recognition dataset.The experimental results show that the proposed GSTEN could achieve an accuracy of 76.54% without increasing the number of model parameters,which was 6.79% higher than the baseline model ST-CGN.

关 键 词：行为识别 骨架序列 全局时空编码网络 猴类动物 药物安全评价 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]