Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, mos...Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, most existing deep learning based recognition frameworks are not optimized for action in the surveillance videos. In this paper, we propose a novel method to deal with the recognition of different types of actions in outdoor surveillance videos. The proposed method first introduces motion compensation to improve the detection of human target. Then, it uses three different types of deep models with single and sequenced images as inputs for the recognition of different types of actions. Finally, predictions from different models are fused with a linear model. Experimental results show that the proposed method works well on the real surveillance videos.展开更多
城市轨道交通起讫点(origin-destination,OD)客流短时预测在智能交通系统中意义重大,它为交通管控策略实施以及出行者出行选择提供了重要的决策依据。卷积神经网络被广泛用于交通数据空间相关性提取,但其平移不变性与局部敏感性导致该...城市轨道交通起讫点(origin-destination,OD)客流短时预测在智能交通系统中意义重大,它为交通管控策略实施以及出行者出行选择提供了重要的决策依据。卷积神经网络被广泛用于交通数据空间相关性提取,但其平移不变性与局部敏感性导致该方法更重视局部特征而忽视全局特征。本研究构建了基于注意力机制的异构数据特征提取机模型(heterogeneous data feature extraction machine,HDFEM)以实现OD矩阵空间相关性的全局感知。该模型从时空特征和用地属性特征出发,构造异构数据OD时空张量与地理信息张量,依托模型张量编码层对异构数据张量进行分割与编码,通过注意力机制连接各张量块特征,提取OD矩阵中各个部分间的空间相关性。该方法不仅实现了异构数据与OD客流数据的融合,还兼顾了卷积神经网络模型未能处理的OD矩阵远距离特征,进而帮助模型更全面地学习OD客流的空间特征。对于OD时序特性,该模型使用了长短时记忆网络来处理。在杭州地铁自动售检票系统(auto fare collection,AFC)数据集上的实验结果表明:HDFEM模型相对于基于卷积神经网络的预测模型,其均方误差、平均绝对误差与标准均方根误差分别下降了4.1%,2.5%,2%,验证了全局OD特征感知对于城市轨道交通OD客流预测的重要性。展开更多
针对汽车生产设备故障领域命名实体识别过程中存在内在语义信息缺失,传统字向量抽取特征单一的问题,提出一种融合部首特征和BERT的汽车生产设备故障领域命名实体识别模型.首先创建汽车生产设备故障领域拆字字典,实现部首特征提取,融合B...针对汽车生产设备故障领域命名实体识别过程中存在内在语义信息缺失,传统字向量抽取特征单一的问题,提出一种融合部首特征和BERT的汽车生产设备故障领域命名实体识别模型.首先创建汽车生产设备故障领域拆字字典,实现部首特征提取,融合BERT预训练模型生成的动态字向量从而得到联合字向量;然后将联合字向量输入到双向长短期记忆(bidirectional long-short term memory,BiLSTM)进行双向编码,获得长序列语义特征;最后通过条件随机场(conditional random field,CRF)进行序列解码,学习标签之间的依赖关系,得到全局最优序列.将该方法在真实汽车生产设备故障领域数据集上进行实验,得到的精确率、召回率、F1值分别为88.4%、90.2%、89.3%.展开更多
文摘Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, most existing deep learning based recognition frameworks are not optimized for action in the surveillance videos. In this paper, we propose a novel method to deal with the recognition of different types of actions in outdoor surveillance videos. The proposed method first introduces motion compensation to improve the detection of human target. Then, it uses three different types of deep models with single and sequenced images as inputs for the recognition of different types of actions. Finally, predictions from different models are fused with a linear model. Experimental results show that the proposed method works well on the real surveillance videos.
文摘城市轨道交通起讫点(origin-destination,OD)客流短时预测在智能交通系统中意义重大,它为交通管控策略实施以及出行者出行选择提供了重要的决策依据。卷积神经网络被广泛用于交通数据空间相关性提取,但其平移不变性与局部敏感性导致该方法更重视局部特征而忽视全局特征。本研究构建了基于注意力机制的异构数据特征提取机模型(heterogeneous data feature extraction machine,HDFEM)以实现OD矩阵空间相关性的全局感知。该模型从时空特征和用地属性特征出发,构造异构数据OD时空张量与地理信息张量,依托模型张量编码层对异构数据张量进行分割与编码,通过注意力机制连接各张量块特征,提取OD矩阵中各个部分间的空间相关性。该方法不仅实现了异构数据与OD客流数据的融合,还兼顾了卷积神经网络模型未能处理的OD矩阵远距离特征,进而帮助模型更全面地学习OD客流的空间特征。对于OD时序特性,该模型使用了长短时记忆网络来处理。在杭州地铁自动售检票系统(auto fare collection,AFC)数据集上的实验结果表明:HDFEM模型相对于基于卷积神经网络的预测模型,其均方误差、平均绝对误差与标准均方根误差分别下降了4.1%,2.5%,2%,验证了全局OD特征感知对于城市轨道交通OD客流预测的重要性。
文摘针对汽车生产设备故障领域命名实体识别过程中存在内在语义信息缺失,传统字向量抽取特征单一的问题,提出一种融合部首特征和BERT的汽车生产设备故障领域命名实体识别模型.首先创建汽车生产设备故障领域拆字字典,实现部首特征提取,融合BERT预训练模型生成的动态字向量从而得到联合字向量;然后将联合字向量输入到双向长短期记忆(bidirectional long-short term memory,BiLSTM)进行双向编码,获得长序列语义特征;最后通过条件随机场(conditional random field,CRF)进行序列解码,学习标签之间的依赖关系,得到全局最优序列.将该方法在真实汽车生产设备故障领域数据集上进行实验,得到的精确率、召回率、F1值分别为88.4%、90.2%、89.3%.
