Network Intrusion Detection based on Improved CNN
Shiyu Wang, Zilong Yang, Anyi Xiao · 2022 5th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE) · 2022
This paper aims to improve the defects of traditional convolutional neural networks in intrusion detection systems and provide an analysis of dynamic data flow. The system first preprocesses the off-line network traffic through numeration, normalization, and grey value mapping. Secondly, the construction method of data visualization processing after the above transformation is explored. Two representative methods are proposed: right-shift filling of a single data cycle and right-shift filling of multiple data cycles. The network traffic captured online is input to obtain the real-time detection results through THE CNN engine interface. Finally, we evaluate the training model. Through comparison with relevant literature and real-time flow detection test, it is found that the optimized visual construction scheme explored has an excellent effect on accuracy and can detect basic flow types in real-time.