-
摘要:
利用人工智能和深度学习技术自动化地分析互联网海量图片,快速、准确地识别有害的暴恐图像并及时处置是反恐工作的重要手段之一。研究了利用深度学习和迁移学习技术对暴恐图像进行分类识别。首先,定义了暴恐图像的主要概念特征,并针对性地构建数据集;其次,针对暴恐图像正样本较少的问题,设计深度神经网络模型和迁移学习方式;最后,基于构建的训练数据集进行模型训练和测试。结果显示:所提方法可以快速、准确地对互联网图片进行分类识别,平均分类准确率达到96.7%,从而有效降低人工检测的劳动强度,为反恐预警工作提供决策支持。
Abstract:Using AI and deep learning technology to automatically analyze massive Internet pictures, quickly and accurately identifying harmful images related to terrorism and dealing with them in time is one of the important means for anti-terrorism work. This paper studies how to use deep learning and transfer learning technology to classify and recognize the images related to terrorism. First, we define the main concept features of the image related to terrorism and collect the relevant positive samples to construct dataset. Second, we design suitable deep neural network model and transfer learning method for the problem of less positive samples of the image related to terrorism. Finally, using the constructed training dataset to fine-tune the final model. The results show that, based on the proposed method in this paper, we can classify and recognize the Internet pictures which have terrorism content quickly and accurately with average classification accuracy rate of 96.7%, and thus the labor intensity of manual monitoring will reduce effectively, which can provide support for decision-making in the work of anti-terrorism early warning.
-
表 1 数据集数量统计
Table 1. Dataset quantity statistics
类别 样本总数量/张 训练数据集/张 验证数据集/张 特定标志类 960 768 192 持枪支或刀者 860 688 172 蓄胡须者 1 200 960 240 特定穿着者 640 512 128 黑布蒙面者 830 664 166 正常图片 5 000 4 000 1 000 表 2 不同模型平均分类准确率及训练收敛时间
Table 2. Average classification accuracy rate and training time of different models
模型 输入图像尺寸/像素 平均分类准确率/% 训练收敛时间/轮 SVM 52.3 densenet121 224 89.6 85 resnext101 224 92.9 120 efficientnet-b3 300 94.2 60 表 3 模型组合的平均分类准确率
Table 3. Average classification accuracy rate of ensemble model
模型组合 平均分类准确率/% efficientnet-b3+densenet121 94.1 efficientnet-b3+resnext101 96.5 resnext101+densenet121 93.2 efficientnet-b3+resnext101+densenet121 96.7 -
[1] 李龙, 支庭荣."算法反恐":恐怖主义媒介化与人工智能应对[J].现代传播(中国传媒大学学报), 2018(9):13-18. http://www.cnki.com.cn/Article/CJFDTotal-XDCB201809002.htmLI L, ZHI T R."Algorithmic anti-terrorism":Terrorism media and the response based on artificial intelligence[J].Modern Communication(Journal of Communication University of China), 2018(9):13-18(in Chinese). http://www.cnki.com.cn/Article/CJFDTotal-XDCB201809002.htm [2] 倪叶舟, 张鹏, 扈翔, 等.大数据背景下暴恐信息挖掘方法综述[J].中国公共安全(学术版), 2018(4):91-95. http://www.cnki.com.cn/Article/CJFDTotal-GGAQ201804020.htmNI Y Z, ZHANG P, HU X, et al.Summarization of the methods of information mining in the background of big data[J].China Public Security(Academy Edition), 2018(4):91-95(in Chinese). http://www.cnki.com.cn/Article/CJFDTotal-GGAQ201804020.htm [3] 符亚彬.基于Logo标志检测的暴恐视频识别系统的设计与实现[D].北京: 北京交通大学, 2016: 15-30. http://cdmd.cnki.com.cn/Article/CDMD-10004-1016115874.htmFU Y B.Design and implementation of the violent-terrorist video recognition system based on Logo markers detection[D].Beijing: Beijing Jiaotong University, 2016: 15-30(in Chinese). http://cdmd.cnki.com.cn/Article/CDMD-10004-1016115874.htm [4] 张宁, 朱金福.机场区域中人群涉暴恐动作智能识别方法仿真[J].计算机仿真, 2015, 32(6):67-70. http://www.cnki.com.cn/Article/CJFDTotal-JSJZ201506016.htmZHANG N, ZHU J F.Intelligent recognition method simulation of ccritical action of people involved in airport areas[J].Computer Simulation, 2015, 32(6):67-70(in Chinese). http://www.cnki.com.cn/Article/CJFDTotal-JSJZ201506016.htm [5] 王胜华.涉暴恐音视频犯罪实证研究——以中国裁判文书网公开的48个判例为分析样本[J].江西警察学院学报, 2019(6):89-96. https://www.zhangqiaokeyan.com/academic-journal-cn_journal-jiangxi-police-institute_thesis/0201275876234.htmlWANG S H.An empirical study on audio and video of violent terrorist crimes-Take 48 cases published by China judicial document network as the analysis sample[J].Journal of Jiangxi Police College, 2019(6):89-96(in Chinese). https://www.zhangqiaokeyan.com/academic-journal-cn_journal-jiangxi-police-institute_thesis/0201275876234.html [6] 黄炜, 黄建桥, 李岳峰.基于BiLSTM-CRF的暴恐信息实体识别模型研究[J].情报杂志, 2019, 38(12):149-156. http://www.cnki.com.cn/Article/CJFDTotal-QBZZ201912022.htmHUANG W, HUANG J Q, LI Y F.Research on entity identification model of terrorism-related information based on BiLSTM-CRF[J].Journal of Intelligence, 2019, 38(12):149-156(in Chinese). http://www.cnki.com.cn/Article/CJFDTotal-QBZZ201912022.htm [7] 廖浚斌, 周欣, 何小海, 等.面向暴恐领域的知识图谱构建方法[J].信息技术与网络安全, 2019, 38(9):34-38. http://www.cnki.com.cn/Article/CJFDTotal-WXJY201909007.htmLIAO J B, ZHOU X, HE X H, et al.Construction method of knowledge graph for terrorism domain[J].Information Technology and Network Security, 2019, 38(9):34-38(in Chinese). http://www.cnki.com.cn/Article/CJFDTotal-WXJY201909007.htm [8] YAN L C, BERNHARD E B, JOHN S D, et al.Backpropagation applied to handwritten zip code recognition[J].Neural Computation, 1989, 1(4):541-551. doi: 10.1162/neco.1989.1.4.541 [9] BENGIO Y, COURVILLE A, VINCENT P.Representation learning:A review and new perspectives[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(8):1798-1828. http://cn.bing.com/academic/profile?id=c85465126b23431b49f9a58196c59bac&encoded=0&v=paper_preview&mkt=zh-cn [10] HINTON G E, SALAKHUTDINOV R R.Reducing the dimensionality of data with neural networks[J].Science, 2006, 313(5786):504-507. doi: 10.1126/science.1127647 [11] KRIZHEVSKY A, SUTSKEVER I, HINTON G E.Imagenet classification with deep convolutional neural networks[C]//Advances in Neural Information Processing Systems.Cambridge: MIT Press, 2012: 1097-1105. https://blog.csdn.net/hongbin_xu/article/details/80271291 [12] HUANG G, LIU Z, WEINBERGER K Q, et al.Densely connected convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway: IEEE Press, 2017: 2261-2269. http://en.cnki.com.cn/Article_en/CJFDTotal-JSJS201810019.htm [13] HU J, SHEN L, SUN G.Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway: IEEE Press, 2018: 7132-7141. https://pubmed.ncbi.nlm.nih.gov/31034408/ [14] TAN M, LE Q.EfficientNet: Rethinking model scaling for convolutional neural networks[C]//Proceedings of the 36th International Conference on Machine Learning.Long Beach: PMLR, 2019: 6105-6114. https://blog.csdn.net/weixin_37993251/article/details/91353858 [15] KINGMA D, BA J.Adam: A method for stochastic optimization[C]//International Conference on Learning Representations, 2014. https://blog.csdn.net/weixin_37993251/article/details/88723271