Abnormal traffic detection is crucial for network security and quality of service. However, the feature similarity and the one-dimensional of the detection model bring great difficulties to the abnormal traffic detection, so a big-step convolutional neural network traffic detection model based on the attention mechanism is proposed. First, the features of the network traffic are analyzed and the raw traffic is pre-processed and mapped to a two dimensional grey scale image. Then the histogram equalization is applied to obtain multi-channel grayscale images, and an attention mechanism is adopted to assign different weights to traffic features to enhance local features. Finally, the pooling-free convolutional neural networks are combined to extract the traffic features of different depths, so as to improve the defects such as local feature omission and overfitting in convolutional neural networks. In the simulation experiment, the public data set and real data set were used to carry out the experiment. The proposed model is compared with ANN, CNN, RF, Bayes and two latest models using the commonly used algorithm SVM as a baseline model. In the experiment with multiple classifications, the accuracy rate is 99.5%. The proposed model has best anomaly detection. And the proposed method outperforms other models in terms of precision, recall and F1. The model is not only efficient in detection, but also robust and robust to different complex environments is demonstrated
REDDY et al. (Fri,) studied this question.