基于DCNN-GRU模型的XSS攻击检测方法  

XSS ATTACK DETECTION METHOD BASED ON DCNN-GRU MODEL

作　　者：许丹丹 徐洋[1] 张思聪[1] 付子爔 Xu Dandan;Xu Yang;Zhang Sicong;Fu Zixi(Key Laboratory of Information and Computing Science of Guizhou Province,Guizhou Normal University,Guiyang 550001,Guizhou,China)

机构地区：[1]贵州师范大学贵州省信息与计算科学重点实验室,贵州贵阳550001

出　　处：《计算机应用与软件》2022年第2期324-329,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61461009);中央引导地方科技发展专项资金项目(黔科中引地[2018]4008);贵州省研究生科研基金立项课题(黔教合YJSCXJH[2019]043);贵州师范大学创新创业教育研究基金项目(SCJJ1805)。

摘　　要：为了提高跨站脚本攻击的检测效率,利用一维DCNN快速处理时序问题的能力和GRU模型处理上下文具有长期依赖关系问题的能力,提出基于DCNN-GRU模型的XSS攻击检测方法。对原始数据做规范化处理,将数据转化为可以对深度学习网络模型进行输入的特征向量。通过卷积层和池化层处理特征向量,GRU层作为门控机制来保留代码间的依赖关系。通过全连接层实现归一化处理,利用Softmax分类器实现分类完成攻击检测。使用foxscheduler数据集进行对比实验,结果表明,DCNN-GRU模型与单一的DCNN、GRU、LSTM模型及SVM模型相比,训练时间更短,检测结果中准确率、召回率和F1值都是最高的。The one-dimensional DCNN can process timing problem quickly,and the GRU model can process context problems which have long time dependence relationship.In order to improve the efficiency of detection of the XSS attack,these two deep learning models are combined based on their ability,and an XSS attack detection method is proposed based on DCNN-GRU model.The raw data was standardized,and it could transform into feature vectors aim to fit the deep learning network models.Feature vectors were processed by the convolutional layer and the pooling layer,and the GRU layer was used as a gating mechanism to preserve the dependencies between the codes.It used the fully connected layer to complete the normalization process,and the attack detection was achieved by using the Softmax classifier to realize classification.Comparative experiments were carried out by using foxscheduler dataset.The experimental results show that the DCNN-GRU model has shorter training time,and its accuracy,recall and F1 values are the highest,compared with the single DCNN,GRU,LSTM models and SVM model in machine learning which verified the detection effect of the DCNN-GRU model.

关 键 词：WEB应用安全 深度学习 跨站脚本 深度卷积神经网络 门控循环单元 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]