Abnormal Traffic Detection Based on Attention and Big Step Convolution

Abnormal traffic detection is critical to network security and quality of service. However, the similarity features single dimension model cause great difficulties for abnormal detection, thus a big-step convolutional neural based on attention mechanism proposed. Firstly, characteristics are analyzed raw preprocessed mapped into two-dimensional grayscale image. Then, multi-channel images generated by histogram equalization, an introduced assign different weights enhance local features. Finally, pooling-free networks combined extract depths, improving defects such as feature omission overfitting in networks. The simulation experiment was carried out balanced public data set actual set. Using commonly used algorithm SVM baseline, proposed compared with ANN, CNN, RF, Bayes two latest models. Experimentally, accuracy rate multiple classifications 99.5%. has best anomaly detection. And method outperforms other models precision, recall, F1. It demonstrated that not only efficient but also robust complex environments.