Abstract: Compared with the previous network data,network data shows more huge,complex and multidimensional characteristics nowadays.In face of the high dimensional data features, traditional machine learning methods need to extract a large number of features manually.Besides,feature extraction process is complex and computational,which is not conducive to the current network intrusion detection real-time and accuracy requirements.Deep learning methods have good advantages in dealing with complex data,which can automatically extract better representation features from the data.In this paper,an intrusion detection method based on deep convolution neural network was proposed.Firstly,a method of transforming network data into images was proposed.Then a deep convolution neural network model was designed for the transformed image,which uses the two-layer convolution layer and the pool layer to reduce the dimension of the image,and introduced the Relu function as a new nonlinear activation in place of the traditional neural network.The sigmoid or Tanh function was used to speed up the convergence of the network,and the Dropout method was introduced in the model to prevent the network model from over-fitting.Finally,the image was trained and identified by constructing the completed depth convolution neural network model.The experimental results show that the proposed method has better detection accuracy,lower false alarm rate and higher detection rate compared with the existing me-thods.

Key words: Convolution neural network, Deep learning, Feature extraction, Intrusion detection, Over-fitting