摘要
传统恶意流量识别方法存在对流量特征区分度不强以及数据集类别不平衡的问题,严重影响模型的识别效果。基于此,根据流量数据的特点,使用一维卷积神经网络提取流量特征,并引入多头注意力机制从多个维度捕捉流量内部不同特征间的依赖性以增强关键特征的区分度,更加全面地提取对恶意流量识别有用的关键特征;对识别模型的损失函数进行设计以解决数据集类别不平衡问题。实验显示,该方法整体识别准确率较现有方法最高提升7%。
Traditional malicious traffic identification methods have the problems of weak discrimination of traffic features and unbalance of data sets,which seriously affect the recognition effect of the model.Based on this,according to the characteristics of traffic data,one-dimensional convolutional neural network is used to extract traffic features,and multi-head attention mechanism is introduced to capture the dependence of different features within traffic from multiple dimensions to enhance the discrimination of key features,so as to extract the key features useful for malicious traffic identification more comprehensively.The loss function of rec-ognition model is designed to solve the problem of class imbalance in data sets.Experimental results show that the overall recognition accuracy of the proposed method is up to 7%higher than that of the existing methods.
作者
郭祥
姜文刚
GUO Xiang;JIANG Wengang(College of Automation,Jiangsu University of Science and Technology,Zhenjiang 212003)
出处
《计算机与数字工程》
2025年第8期2228-2233,共6页
Computer & Digital Engineering
基金
国家自然科学基金项目(编号:61702235)
江苏省研究生创新计划(编号:KYCX21_3482)资助。
关键词
恶意流量
深度学习
多头注意力机制
卷积神经网络
类别不平衡
malware traffic
deep learning
multi-head attention mechanism
convolutional neural network(CNN)
class imbalance
