MalEdgeNeXt: A Lightweight Malware Family Classification Method
Zepeng Wu, Wenyin Yang, Linjiu Guo, Yuan Dong, Ziliang Liu, Li Ma · 2023
The industrial internet has been compromised by malicious software in recent years. However, malicious software detection and analysis models are difficult to run in industrial equipment, due to the limitation of computing resources. Therefore, research of lightweight classification methods and analysis of malware families has become an important issue in industrial internet security. Traditional malicious software data analysis methods face challenges such as complex labeling, long processing time when dealing with massive data, and the real-time requirements of malicious software classification in the industrial Internet environment. As we know, deep learning performs well in image processing, can automatically extract image feature information and classify. Inspired of this, this paper proposes a lightweight malware family classification model MalEdgeNeXt. It converts the binary files of malicious code into images, whose features would be extracted by the lightweight EdgeNeXt model with a customized Linear+softmax classifier. The experimental results show that MalEdgeNeXt has higher classification accuracy and lower parameters than the compared lightweight network models. Therefore, MalEdgeNeXt is suitable for industrial control systems, and provides security guarantees for industrial internet.