摘要
面对复杂场景下异常事件检测的准确率偏低的情况,本文提出一种基于深度学习的异常事件检测方法,并将此方法扩展为异常事件分类方法.利用神经网络模型提取特征,将群体发散聚集事件,群体密集聚集事件,群体逃散事件和追赶事件这4种异常事件进行检测和分类.通过PKU-SVD-B测试集对训练出来的模型进行测试实验,并在UMN数据集上与几种方法做了对比实验,验证了本文提出的基于深度学习的异常事件检测算法,在适应多种不同场景的前提下,对多种异常事件检测的准确率很高,表明训练出来的模型对异常事件检测具有极强的泛化能力.
Faced with low accuracy of abnormal event detection in complex scenarios,this paper proposes an abnor-mal event detection based on deep learning in various public scenes and multiple types of anomalies,and the method has been extended to an abnormal event classification method.The neural network model is used to extract features,and the four kinds of abnormal events,such as group divergence aggregation events,group intensive aggregation events,group escape e-vents and catch-up events,are detected and classified.Test the trained model with PKU-SVD-B test set,compared with vari-ous methods on the UMN datasets,and verify the algorithm of abnormal event detection based on deep learning proposed in this paper.Under the premise of adapting to different scenarios,various abnormal events are detected.The high accuracy rate indicates that the trained model has strong ability to generalize abnormal event detection.
作者
闻佳
王宏君
邓佳
刘鹏飞
WEN Jia;WANG Hong-jun;DENG Jia;LIU Peng-fei(School of Information Science and Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;The Key Laboratory for Computer Virtual Technology and System Integration of Hebei Province,Qinhuangdao,Hebei 066004,China;State Key Laboratory of Software Engineering of Hebei Province,Qinhuangdao,Hebei 066004,China)
出处
《电子学报》
EI
CAS
CSCD
北大核心
2020年第2期308-313,共6页
Acta Electronica Sinica
关键词
异常检测
异常分类
深度学习
图像处理
卷积神经网络
特征提取
abnormal detection
abnormal classification
deep learning
image processing
convolutional neural net-work
feature extraction