基于多尺度融合和时空特征的网络入侵检测模型
Network intrusion detection model based on multiscale fusion and spatial-temporal features
龚星宇 1来源 1李娜 1雷璇1
作者信息
- 1. 西安科技大学计算机科学与技术学院,陕西西安 710600
- 折叠
摘要
针对入侵检测模型提取特征能力不足,且流量数据中含冗余噪声的问题,提出一种基于多尺度融合和时空特征的ML-PFN入侵检测模型.采用多尺度特征融合技术分别提取数据中浅层特征信息和深层特征信息,使模型学习的特征更加丰富;采用软阈值函数和注意力机制自动选择合适的阈值,减少噪声及不相关信息对模型的干扰;融合时空特征构成多尺度空间特征提取长短时记忆-并行特征网络(MSFE LSTM-parallel feature network,ML-PFN)模型,并应用于网络入侵检测.通过3个公开数据集进行性能评估,实验结果表明,ML-PFN模型对比其它5种分类模型各项指标效果最好,在训练时长适中的同时准确率达到96.45％.
Abstract
Aiming at the problems that the intrusion detection model lacks the ability to extract features and the traffic data con-tains redundant noise,a ML-PFN intrusion detection model based on multiscale fusion and spatial-temporal features was pro-posed.The multiscale feature fusion technology was used to extract the shallow feature information and the deep feature informa-tion in the data respectively,so as to enrich the features of model learning.Soft threshold function and attention mechanism were used to automatically select appropriate threshold to reduce the interference of noise and irrelevant information on the model.The multiscale spatial feature extraction parallel feature network(ML-PFN)model based on the fusion of spatial-temporal features was constructed and applied to network intrusion detection.Through performance evaluation on three open data sets,the experi-mental results show that ML-PFN model has the best performance compared with other five classification models,and the accu-racy rate reaches 96.45％when the training time is moderate.
关键词
入侵检测/冗余噪声/多尺度融合/时空特征/软阈值/注意力机制/长短时记忆Key words
intrusion detection/redundant noise/multiscale fusion/spatial-temporal features/soft threshold/attention mecha-nism/long and short-term memory引用本文复制引用
出版年
2024
NETL
NSTL
万方数据