Robust detection for network intrusion of industrial IoT based on multi-CNN fusion

Abstract A robust intrusion detection system plays a very important role in network security. In the face of complex network data and diverse intrusion methods, traditional machine learning methods seem to be inadequate and cannot meet the requirements of the current network environment. Existing deep learning-based methods are far from fully exploiting their potential in dealing with such one-dimensional feature data, and their performance is still unsatisfactory in detecting unknown intrusions. This paper proposes a deep learning approach for intrusion detection using a multi-convolutional neural network (multi-CNN) fusion method. According to the correlation, the feature data are divided into four parts, and then the one-dimensional feature data are converted into a grayscale graph. By using the flow data visualization method, CNN is introduced into the intrusion detection problem and the best of the four results emerge. The experimental results successfully demonstrate that the multi-CNN fusion model is very suitable for providing a classification method with high accuracy and low complexity on the NSL-KDD dataset. Furthermore, its performance is also superior to those of traditional machine learning methods and other recent deep learning approaches for binary classification and multiclass classification. This work will contribute for the data security of industrial IoT.