Traffic Image Based Deep Learning Intrusion Detection Method

The continuous popularization and development of Internet technology has made society more and more dependent on the Internet.The Internet has become an important infrastructure for economic and social development.While bringing convenience to the Internet,it also brings security threats to society and individuals.Network intrusion detection technology detects and identifies potential intrusion attacks in network traffic data,provides protection for society and individuals to maintain network security,and resists network security threats.Traditional network intrusion detection technology needs to rely on expert knowledge to manually design a feature library to perform feature matching on network data,or use machine learning-based classification and clustering algorithms to perform feature learning to achieve detection and recognition tasks.The effect of the above method relies heavily on the quality of feature selection design and still lacks a certain "intelligence".Therefore,based on previous research work,this dissertation uses deep learning methods to improve network intrusion detection technology.Using deep learning technology can realize the advantages of automatic feature extraction,directly process network traffic data,image network data,and use the neural network performs feature learning to avoid manually designing traffic data features.Aiming at the task of Web intrusion detection,this dissertation proposes a web intrusion detection method based on attention mechanism.One-hot encoding of network data is used to reconstruct a two-dimensional image,and a neural network model based on attention mechanism is constructed for feature learning.Through the attention mechanism,the attention probability of each part of the data is calculated,and the final feature vector is weighted and summed.The softmax classifier is used to complete the recognition and detection task.And use Dropout technology and Gelu activation function to optimize the model to prevent the model from falling into overfitting.This dissertation also proposes an intrusion detection method based on traffic feature images.For network traffic data,the data is mapped to the RGB value space to generate a three-channel traffic image.At the same time,the information entropy of the network data is added to the RGB traffic image as transparency to generate.RGBA four-channel flow image,construct a convolutional neural network to extract features from the generated flow image,and realize the feature learning of the flow data.In view of the imbalance problem in network data,the Focal loss function is introduced to improve the detection effect of the model on unbalanced data sets.Through qualitative and quantitative comparison experiments,the above two methods can achieve better detection results in their respective detection tasks,and the classification accuracy is above 95%.The thesis aims at the network intrusion detection task,through the feature extraction and classification of the abnormal data in the network to realize the identification and recognition of the abnormal intrusion behavior in the network traffic data.The experimental results prove that the method proposed in the dissertation is feasible.