基于特征增强的深度学习入侵检测算法

Deep Learning Intrusion Detection Algorithm Based on Feature Enhancement

深度学习技术随着社会的不断发展,逐渐应用到了社会生活的各个领域,其中与入侵检测技术相结合已成为当今的研究热点。在当前不稳定的网络环境下,能够准确识别异常流量是当前入侵检测的主要任务。传统的神经网络面对复杂的数据,无法准确提取到与分类结果有关的特征,过多的冗余特征会导致模型泛化能力差,检测准确率不高,这会导致深度学习技术无法很好地应用在入侵检测的任务中。为了解决这个问题,提出了一种基于特征增强的深度学习入侵检测方法,即在模型训练过程中,通过一个辅助网络,增强对分类结果相关的特征,使模型着重学习对分类有益的特征,同时减轻冗余特征对模型分类的影响。同时,该方法不会修改原有模型的结构,可以轻松地应用在不同的卷积神经网络的模型上。最后在NSL-KDD和CICIDS2017数据集上的实验结果表明,准确率最高可达99.73%和99.15%。

Deep learning technology has been gradually applied to various fields of social life with the continuous development of society,among which combining with intrusion detection technology has become a hot research topic today.In the current unstable network environment,accurate identification of abnormal traffic is the main task of current intrusion detection.Traditional neural networks cannot accurately extract features related to classification results in the face of complex data,and too many redundant features will lead to poor generalization of the model and low detection accuracy,which will lead to the inability of deep learning techniques to be well applied in the task of intrusion detection.To solve this problem,we propose a deep learning intrusion detection method based on feature enhancement.During the model training process,the features that are relevant to the classification results are enhanced by an auxiliary network,so that the model focuses on learning features that are beneficial to the classification,while mitigating the impact of redundant features on the model classification.At the same time,the proposed method does not modify the structure of the original model and can be easily applied to different models of convolutional neural networks.Finally,experiments on the NSL-KDD and CICIDS2017 datasets show that the accuracy is up to 99.73%and 99.15%.