Related Experiment Video
Updated: Nov 6, 2025
08:20
Author Spotlight: AI-Driven Trypanosome Species Detection from Microscopic Images
Published on: October 27, 2023
2.0K
Malware homology determination using visualized images and feature fusion
Xuejin Zhu1, Jie Huang1,2, Bin Wang3
1School of Cyber Science and Engineering, Southeast University, Nanjing, Jiangsu, China.
Peerj. Computer Science
|May 12, 2021
Summary
This study introduces a novel malware homology determination method by combining global and local features. The new approach significantly improves malware family classification accuracy compared to existing methods.
Area of Science:
- Computer Science
- Cybersecurity
- Machine Learning
Background:
- Malware variants are increasing, making family homology determination crucial.
- Existing methods rely on global features, allowing intentional misclassification by malware creators.
Purpose of the Study:
- To develop an improved malware homology determination method using fused global and local features.
- To enhance the accuracy and robustness of malware family classification.
Main Methods:
- Converted malware executable global structure into bytecode images.
- Extracted opcode semantic information using n-gram models to generate opcode images.
- Proposed a dual-branch convolutional neural network utilizing both image types for classification.
Main Results:
- Achieved 99.05% accuracy and 98.52% F-measure in malware family homology classification.
- Outperformed existing schemes relying on single image features.
Conclusions:
- The proposed method effectively fuses global and local features for superior malware classification.
- This approach offers a more reliable solution for identifying malware families amidst rising variants.
