Research on Network Intrusion Detection Technology Based on DeepInsight and Transfer Learning

For the problems such as less intrusion training samples and unbalanced samples are existing in intrusion detection research, this paper proposes an intrusion detection method based on DeepInsight and Transfer learning DI-TL-CNN (DI-TL-CNN). The intrusion detection method uses DeepInsight method to preprocess the intrusion data set into the image data set suitable for CNN model input. VGG16 model in CNN model is selected as the basic learning model, and transfer learning transfers the pre-trained CNN model to the current task. By freezing and fine-tuning different module parameters in the CNN model, four transfer schemes are proposed. Experiments were carried out on UNSW-NB15 network intrusion dataset, and the results of four migration schemes showed that the DI-TL-CNN model fine-tuning method combined with transfer learning, the more parameters fine-tuned, the stronger the ability of the model to extract top-level features, and the better the comprehensive performance of the model. The experimental results show that the method based on DI-TL-CNN model is superior to other detection methods in accuracy and performance, and has a good application prospect.