基于卷积循环的知识蒸馏入侵检测方法OA

Aiming at the problems that existing intrusion detection methods based on deep learning have insufficient ability to extract mixed features of traffic data and low efficiency of model deployment,this paper proposed an intrusion detection model combining convolutional cyclic network and knowledge distillation.This model uses the convolution cycle structure to extract the sequential mixture features of traffic data,and proposes a two-stage knowledge distillation method for the convolution cycle structure.The attention mechanism is introduced in the process of feature distilling students network structure to optimize the initial parameters,and then introduce relevance in the distillation loss to transfer the middle layer relationship knowledge of the teacher network and fine-tune the global parameters.The experimental results show that the proposed model has better deployment efficiency than the traditional intrusion model,and the proposed method improves the performance of the original model by 5.49％.