基于深度卷积神经网络的异常行为快速识别  

Fast Detection of Abnormal Behavior Based on Deep Convolutional Neural Network

作　　者：龙翔[1,2] 韩兰胜[1,3] 王伟豪[1] LONG Xiang;HAN Lansheng;WANG Weihao(School of Cyber Space Security,Huazhong University of Science and Technology,Wuhan 430074,China;Hubei Vocational College of Bio-Technology,Wuhan 430070,China;Cyberspace Security Research Center,Peng Cheng Laboratory,Shenzhen 518000,China)

机构地区：[1]华中科技大学网络空间安全学院,武汉430074 [2]湖北生物科技职业学院,武汉430070 [3]鹏程实验室网络空间安全研究中心,广东深圳518000

出　　处：《火力与指挥控制》2023年第1期26-32,共7页Fire Control & Command Control

基　　金：国家自然科学基金资助项目(61272033,62072200,6217071437)。

摘　　要：针对异常行为快速识别问题,提出了一种基于深度卷积神经网络的视频检测和定位方法。该方法利用全卷积神经网络和时间数据,将一个预先经过训练和监督的全卷积神经网络转移到一个无监督的全卷积神经网络,确保能够检测全局场景中的异常,提出利用级联检测的方式来降低算法的计算复杂度,从而使其在速度和精度方面获得较高的性能。提出的基于全卷积神经网络的异常行为检测架构解决了两个主要任务,即特征表示和级联离群值检测。实验结果表明,所提方法在检测和定位精度上优于现有算法,且运行速度更快,从而表明所提算法的有效性和可行性。Aiming at the problem of rapid detection of abnormal behavior a video detection and localization method based on deep convolutional neural network is proposed.This method firstly uses the fully convolutional neural network and time data to transfer a pre-trained and supervised fully convolutional neural network to an unsupervised fully convolutional neural network to ensure that anomalies in the global scene can be detected,and then the cascade detection mode is proposed to be used to reduce the computational complexity of the algorithm,so that the higher performance in terms of speed and accuracy can be obtained.The abnormal behavior detection architecture based on the fully convolutional neural network proposed in this paper solves two main tasks,namely feature representation and cascaded outlier detection.The experimental results show that the method proposed in this paper is superior to the existing algorithms in detection and positioning accuracy,and it runs faster,thus the effectiveness and feasibility of the algorithm proposed in this paper is demonstrated.

关 键 词：异常检测 卷积神经网络 拥挤场景 快速识别 异常行为 

分 类 号：TN953[电子电信—信号与信息处理]