结合深度神经网络的网络监控系统微表情识别  

MICRO-EXPRESSION RECOGNITION FOR NETWORKS MONITORING SYSTEM COMBINED WITH DEEP NEURAL NETWORKS

作　　者：罗予东[1] 李振坤[2] Luo Yudong;Li Zhenkun(School of Computer Technology,Jiaying University,Meizhou 514015,Guangdong,China;School of Computers,Guangdong University of Technology,Guangzhou 510006,Guangdong,China)

机构地区：[1]嘉应学院计算机学院,广东梅州514015 [2]广东工业大学计算机学院,广东广州510006

出　　处：《计算机应用与软件》2023年第7期172-179,191,共9页Computer Applications and Software

基　　金：国家自然科学基金面上项目(41172028)。

摘　　要：为了提高网络监控系统微表情识别的效果,结合深度神经网络技术提出新的智能监控系统微表情识别算法。将判别能力强的卷积神经网络特征与鲁棒的直方图特征结合,利用卷积神经网络提取目标的空间特征,再将卷积特征表示为直方图,结合直方图和卷积神经网络两者的优势设计新的人体追踪算法;设计跨模态监督的深度神经网络训练方法,将可见光视频数据送入深度神经网络进行训练,利用近红外光视频对训练程序进行监督。基于公开的多模态微表情识别数据集完成了验证实验,结果显示该算法有效地提高了微表情识别的性能。In order to improve the micro-expression recognition performance for networks monitoring system,a new micro-expression recognition algorithm for intelligent monitoring system combined with deep neural networks technique is proposed.The algorithm combined strong discriminative convolution neural network features with robust histogram features,and we used convolutional neural networks to extract the objective space features.The convolutional features were represented as histograms,and we took advantages of the strengths of both histograms and convolutional neural networks to design a new human tracking algorithm.A cross-modality supervision training method for deep neural networks was designed,the visible light videos were delivered to deep neural networks to train,and near infrared spectrum videos were used to supervise the training procedure.Validation experiments were carried out on a public multiple modality micro-expression recognition dataset.The results show that the proposed algorithm improves the performance of micro-expression recognition.

关 键 词：卷积神经网络 智能监控系统 微表情识别 深度神经网络 跨模态融合 近红外光成像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]