中图分类号：TN918；TP183 文献标志码：A DOI: 10.16157/j.issn.0258-7998.244838
中文引用格式： 丁全，丁伯瑞，查正朋，等. 恶意代码可视化分类研究[J]. 电子技术应用，2024，50(5)：41-46.
英文引用格式： Ding Quan，Ding Borui，Zha Zhengpeng，et al. Research on visualization-based classification of malicious code[J]. Application of Electronic Technique，2024，50(5)：41-46.

Research on visualization-based classification of malicious code

Abstract： The design of new malicious code is becoming increasingly complex, and traditional recognition and detection methods can no longer meet current requirements. Therefore, based on the analysis of the BODMAS dataset, this paper performs visualization processing and classification. At the same time, considering that the existing malware visualization classification models mainly rely on global features, this paper designs a CA (Channel-level local feature Attention) module and a MA (Multi-scale local feature Attention) module based on the convolutional neural network, and constructs two new models that cleverly combine global and local features. On the BODMAS dataset, the new models have achieved an increase in the average accuracy of recognizing and classifying malware types compared to the methods described in the BODMAS dataset paper. This proves the feasibility of dataset visualization and the effectiveness of the new models, providing important data and experimental basis for future research.

Key words : BODMAS dataset；CA module；MA module；visualization of malicious code