计算机科学

计算机科学 ›› 2024, Vol. 51 ›› Issue (6A): 231000052-6.doi: 10.11896/jsjkx.231000052

• 信息安全 • 上一篇    下一篇

基于两阶段算法的多媒体有害信息识别方法

史晓苏1,2, 李欣1, 简玲2, 倪华健3   

  1. 1 中国人民公安大学信息网络安全学院 北京 100091
    2 上海市公安局网络安全保卫总队 上海 200025
    3 上海闪马智能科技有限公司 杭州 310000
  • 发布日期:2024-06-06
  • 通讯作者: 李欣(lixin@ppsuc.edu.cn)
  • 作者简介:(903323721@qq.com)
  • 基金资助:
    公安部应用创新计划(2020YYCXSHSJ019)

Multimedia Harmful Information Recognition Method Based on Two-stage Algorithm

SHI Xiaosu1,2, LI Xin1, JIAN Ling2, NI Huajian3   

  1. 1 Information Network Security Academy,People’s Public Security University of China,Beijing 100091,China
    2 Network Security Corps,Shanghai Public Security Bureau,Shanghai 200025,China
    3 Shanghai SUPREMIND Technology Co.,Ltd,Zhejiang 310000,China
  • Published:2024-06-06
  • About author:SHI Xiaosu,born in 1996,postgraduate.Her main research interests include ima-ge processing and big data analysis.
    LI Xin,born in 1977,Ph.D,associate professor,is a professional member of CCF(No.51691M).His main research interests include cyber security and big data analysis.
  • Supported by:
    Applied Innovation Plan of the Ministry of Public Security of China(2020YYCXSHSJ019).

PDF (PC)

321

摘要： 在互联网安全监管和网络违法犯罪打击整治的应用场景中,现有多媒体有害信息识别方法普遍存在运算效率不高、无法准确识别局部敏感信息,以及识别检测局限于单一的网络违法犯罪类型等问题。针对以上问题,文中提出了一种基于两阶段算法的多媒体有害信息识别模型。该模型将信息过滤与内容检测分阶段处理,将场景识别和元素目标检测分任务并行处理,第一阶段采用EfficientNet-B2构建高吞吐的前置过滤模块快速筛选掉80%正常内容的数据;第二阶段基于Meal-V2,Faster RCNN,NetVLAD网络构建3种不同网络结构的模块,适应多维度场景、多特征元素的识别要求。结果表明,模型运算效率在T4卡上达到57FPS,多媒体有害信息的识别准确率、召回率均超过97%;与传统模型相比,在NPDI和自建测试集上识别准确率分别最高提升3.09%和19.26%。

关键词: 两阶段算法, 多媒体, 有害信息识别

Abstract: In the application scenarios of Internet content security supervision and combating and rectifying Internet crimes,existing multimedia harmful information identification methods generally have problems such as low computational efficiency,inability to accurately identify local sensitive information,and identification capabilities are limited to a single type of cyber crimes.In order to solve the above problems,the paper proposes a multimedia harmful information recognition model based on a two-stage algorithm.This method processes information filtering and content detection in stages,and splits the tasks of scene recognition and element target detection.The first stage uses EfficientNet-B2 to build a high-throughput pre-filter model to quickly filter out 80% of images and short videos with normal content.In the second stage,three modules with different network structures are built based on Meal-V2,Faster RCNN,and NetVLAD networks to adapt to the recognition requirements of multi-dimensional scenes and multi-feature elements.The results show that the model’s computing efficiency reaches 57FPS(frames per second) on the T4 card,and the recognition accuracy and recall rate of multimedia harmful information exceed 97%.Compared with traditional mo-dels,the recognition accuracy rate on the NPDI dataset and the self-built test dataset increases by 3.09% and 19.26% respectively.

Key words: Two-stage algorithm, Multimedia, Harmful information recognition

中图分类号: 

  • TP391.41
[1]HUANG Y,KONG A W K.Using a CNN ensemble for detecting pornographic and upskirt images[C]//2016 IEEE 8th International Conference on Biometrics Theory,Applications and Systems(BTAS).IEEE,2016:1-7.
[2]CONNIE T,AL-SHABI M,GOH M.Smart content recognition from images using a mixture of convolutional neural networks[M]//IT Convergence and Security 2017:Volume 1.Singapore:Springer Singapore,2017:11-18.
[3]MOUSTAFA M.Applying deep learning to classify pornographicimages and videos[J].arXiv:1511.08899,2015.
[4]XIE X.Research on detection technology of pornographic information based on deep learning[D].Chengdu:University of Electronic Science and Technology of China,2020.
[5]ZHANG D L.Research on special video content detection algorithm based on deep features[D].Beijing:Minzu University of China,2019.
[6]GU Y,LI J,JING B,et al.Internet content information detection and filtration system[J].Application Research of Computers,2008(9):2834-2835,2862.
[7]HE K,ZHANG X,REN S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:770-778.
[8]TAN M,LE Q.Efficientnet:Rethinking model scaling for convolutional neural networks[C]//International Conference on Machine Learning.PMLR,2019:6105-6114.
[9]GOU J,YU B,MAYBANK S J,et al.Knowledge distillation:A survey[J].International Journal of Computer Vision,2021,129:1789-1819.
[10]TOUVRON H,CORD M,DOUZE M,et al.Training data-efficient image transformers & distillation through attention[C]//International Conference on Machine Learning.PMLR,2021:10347-10357.
[11]TOUVRON H,VEDALDI A,DOUZE M,et al.Fixing thetrain-test resolution discrepancy[C]//Proceedings of the 33rd International Conference on Neural Information Processing Systems.2019:8252-8262.
[12]MAHAJAN D,GIRSHICK R,RAMANATHAN V,et al.Exploring the limits of weakly supervised pretraining[C]//Proceedings of the European Conference on Computer Vision(ECCV).2018:181-196.
[13]CAI H,ZHU L,HAN S.Proxylessnas:Direct neural architecture search on target task and hardware[J].arXiv:1812.00332,2018.
[14]SHEN Z,SAVVIDES M.Meal v2:Boosting vanilla resnet-50 to 80%+ top-1 accuracy on imagenet without tricks[J].arXiv:2009.08453,2020.
[15]REN S,HE K,GIRSHICK R,et al.Faster r-cnn:Towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.
[16]SHAFIQ M,GU Z.Deep residual learning for image recognition:a survey[J].Applied Sciences,2022,12(18):8972.
[17]DONG S.The research on object detection algorithm based on improved SSD and FPN algorithm[D].Chongqing:Southwest University,2023.
[18]ZHANG Z H.Study on scene recognition algorithm based on NetVLAD[D].Chongqing:Chongqing University,2022.
[19]ARANDJELOVIC R,GRONAT P,TORII A,et al.NetVLAD:CNN architecture for weakly supervised place recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:5297-5307.
[20]AVILA S,THOME N,CORD M,et al.Pooling in image representation:The visual codeword point of view[J].Computer Vision and Image Understanding,2013,117(5):453-446.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发