Design Of Network Steganalysis Model Based On Neural Network

Network steganography mainly uses redundant fields and protocol rules to set loopholes for secret information transmission,which is dynamic and more concealed.Network steganalysis performs compliance testing to identify and destroy secret communications for steganography based on network protocols in the network environment.Existing network steganalysis mainly employed statistical steganography analysis,which has the problems of difficulty in feature extraction and is suitable for detection of one targeted steganography algorithm.To solve these problems,this paper studies the existing network steganography and network steganalysis methods,and constructs some neural network models to achieve network steganography feature extraction and classification.The specific work is as follows:1.A network steganography analysis model BNS-CNN is proposed based on convolutional neural network in IPv6 network.Considering the characteristics of steganography based on IP,TCP and UDP protocol headers in the IPv6 network,the model preprocesses the network data stream and divides the features according to the fields;leverages multiple convolution kernels with different heights for feature extraction and adjusts the width of the convolution kernel to ensure that the field is the smallest unit of feature recognition;K-max pooling is employed to retain more feature information and enhance the intensity of steganographic features.The BNS-CNN model can identify a variety of storage hidden channels based on IPv6 packet headers,and the detection accuracy rate is as high as 99.98%.2.A network steganalysis model DH-NS is proposed based on DenseBlock and heterogeneous convolution.The model preprocesses network data packets into a matrix with the same size,combined with DenseBlock and heterogeneous convolution for steganographic feature enhancement and feature extraction.The model eliminates the pooling layer and uses global convolution to generate feature single values to reduce the loss of steganographic information during the pooling process.The detection accuracy of the DH-NS model is 95.25%.Model detection accuracy is high and time-consuming is reduced.The model has good performance in the field of network storage hidden channel detection,and overcomes the problem that the BNS-CNN model only performs steganalysis for specific protocols,and enhances the universality of network steganalysis.3.A storage network steganalysis model CGNS is proposed based on spatiotemporal features.The model first employs the convolutional neural network to learn the low-level spatial characteristics of network traffic,and then uses the GRU to learn the high-level temporal inter-packet dependence characteristics.The model performs network steganalysis by analyzing individual packet feature changes and implicit feature changes between packets.The detection accuracy of the storage network steganography algorithm is 98.53%,which proves its effectiveness in feature learning.In summary,this paper specifically proposes three network steganalysis models based on neural networks,which avoids the difficulty of manual feature extraction and the single application of the algorithm.These models improve the accuracy,versatility and detection efficiency of network steganalysis.