Malicious Classification Based on Deep Learning and Visualization
WANG Jun-ling, WANG Shuo-hao · 2019
In order to classify a large number of malicious codes quickly and accurately, a malicious code classification model based on deep learning and visualization is proposed. Varying from previous studies, most of them are based on byte features. At First, image processing technology is used to visualize malicious code, in the second stage, a malicious code image classification network (MICN) model is designed to extract image features and classify them, so as to improve the robustness of the method. Preliminary experimental results are quite promising with 95.44% classification accuracy on a malware database of 21,741 samples with 9 different malware families. The experimental results show that this method can effectively classify malicious code and has considerable recognition ability for the obfuscation technique.