基于多频特征学习的恶意代码变种分类
Classification of malicious code variants based on multi-frequency feature learning
靳黎忠 1薛慧琴 2段明博 2赵旭俊 2高改梅2
作者信息
- 1. 太原科技大学应用科学学院,山西 太原 030024
- 2. 太原科技大学计算机科学与技术学院,山西太原 030024
- 折叠
摘要
针对恶意代码变种分类方法没有充分对原始输入进行分析的问题,提出一种更加高效的基于深度学习的办法,使用卷积神经网络对多频信息进行学习.对恶意代码转化而成的图像进行研究,利用小波变换进行多频和多层次的分析,抓住低频和高频特征;针对多频信息输入,设计一种多频特征学习模块,充分挖掘其中有用信息.实验结果表明,该方法在Malimg数据集上,相比其它两种恶意代码分类办法,分别取得了 1.5％和0.8％的效果提升.
Abstract
Aiming at the problems of the insufficient analysis of raw input in malicious code variant classification methods,a more efficient methodology of deep learning was presented.The convolutional neural networks(CNN)was used to learn multi fre-quency information.For the image transformed by malicious code,the wavelet transform was used to carry out multi-frequency and multi-level analysis,and the low frequency and high frequency characteristics were grasped.For the multi-frequency infor-mation input,a multi-frequency feature learning module was designed to fully explore the useful information.Experimental re-sults show that compared with the other two malicious code classification methods,the proposed method achieves 1.5％and 0.8％improvement respectively on Malimg dataset.
关键词
恶意代码分类/多频特征/深度学习/小波变换/灰度图像/卷积神经网络/恶意代码家族Key words
malicious code classification/multi frequency feature/deep learning/wavelet transform/gray scale/convolutional neural network/malicious code family引用本文复制引用
基金项目
国防科技重点实验室基金项目(JSY6142219202114)
太原科技大学科研启动基金项目(20212033)
出版年
2024
NETL
NSTL
万方数据