您的位置：首页 > 论文页面
基于改进CBAM多分支平滑空洞卷积的入侵检测算法
发表时间：2023-06-30 浏览量：529 下载量：79
| 全部作者： | 孙志超，杜晔，黎妹红 |
| 作者单位： | 智能交通数据安全与隐私保护技术北京市重点实验室，北京交通大学计算机与信息技术学院 |
| 摘 要： | 针对现有入侵检测算法模型存在表征能力弱、预测准确率低和漏报率高的问题，本文提出基于改进卷积注意力模块的多分支平滑空洞卷积神经网络（multi-branch smoothed dilation convolutional neural network based on improved convolutional block attention module，MSDCNN-ICBAM) 算法模型。该模型首先使用加权随机采样解决数据集不平衡问题，然后设计扁平式的多分支平滑空洞卷积神经网络（MSDCNN）在多感受野下提取多尺度特征以解决网络退化问题，最后提出改进卷积注意力模块（ICBAM）在通道和空间双维度上指导特征表达以解决卷积操作无法感知特征重要性问题。与其他入侵检测模型在UNSW-NB15数据集上的对比实验表明，该模型准确率提高了3.04%，漏报率降低了5.77%，检测率可达90.98%。 |
| 关 键 词： | 计算机科学技术基础学科；入侵检测算法；平滑空洞卷积；卷积注意力模块；加权随机采样 |
| Title： | An intrusion detection algorithm based on improved CBAM multi-branch smoothed dilation convolution |
| Author： | SUN Zhichao, DU Ye, LI Meihong |
| Organization： | Beijing Key Laboratory of Security and Privacy in Intelligent Transportation, School of Computer and Information Technology, Beijing Jiaotong University |
| Abstract： | There are weak characterization ability, low prediction accuracy and high leakage rate in existing intrusion detection algorithm models. This paper proposed a multi-branch smoothed dilation convolutional neural network based on improved convolutional block attention module (MSDCNN-ICBAM) to solve them. The model firstly used weighted random sampling to solve the imbalance problem of the data set. Then it designed a flat multi-branch smoothed dilation convolutional neural network (MSDCNN) to extract multi-scale features under multiple sensory fields to solve the network degradation problem. Finally it proposed an improved convolutional attention module (ICBAM) to guide the feature representation in both channel and space dimensions to solve the problem that the importance of features couldn’t be sensed by convolutional operation. The comparison experiments with other intrusion detection models on the UNSW-NB15 dataset show that the accuracy of the model is improved by 3.04%, that the leakage rate is reduced by 5.77%, and that the detection rate can reach 90.98%. |
| Key words： | basic subject of science and technology for computer; intrusion detection algorithm; smoothed dilation convolution; convolutional block attention module; weighted random sampling |
| 发表期数： | 2023年6月第2期 |
| 引用格式： | 孙志超，杜晔，黎妹红. 基于改进CBAM多分支平滑空洞卷积的入侵检测算法[J]. 中国科技论文在线精品论文，2023，16（2）：209-222. |
1
评论数 0请您登录
暂无评论
下载全文
多维论文