基于可视化特征的Android恶意软件分类模型

Android Malware Classification Model Based on Visual Features

Android恶意软件严重影响用户的使用体验,威胁用户的财产安全、个人隐私。如何快速且准确地实现Android设备上恶意软件的分类成为研究热点。本文分析Android恶意软件的静态特征,通过可视化方法获得恶意软件的特征图像,采用卷积神经网络(CNN)来学习恶意软件的泛化特征。针对卷积神经网络在特征学习中内存资源占用过大的问题,运用MobileNet-V1模型对卷积神经网络进行轻量化改造,实现Android设备上恶意软件的快速分类。通过实验,本文搭建的模型对Android设备上恶意软件分类准确率达到85.9%,分类速度达到21.95 ms/次。相较于传统方法,本文模型在保持较高分类准确率的情况下大幅提升分类速度,减少网络计算复杂度,可以实现对终端设备上恶意软件的快速且准确的分类。

Android malware will seriously affect the user’s experience,threaten the user’s property safety,personal privacy,etc.Therefore,how to quickly and accurately classify malware on Android devices has become a research hotspot.This paper first analyzes the static features of Android malware,uses visualization methods to obtain feature images of malware,and then uses convolutional neural networks(CNN)to learn the generalized features of malware.Aiming at the problem of convolutional neural network occupying too much memory resources in feature learning,the MobileNet-V1 model is used to light-weight the convolutional neural network to achieve rapid classification of malware on Android devices.Through experiments,the model built in this article has an accuracy of 85.9%for malware classification on Android devices,and the speed of each classification reaches 21.95 ms.Compared with traditional methods while maintaining classification accuracy,the classification speed is greatly improved and the network calculation complexity is reduced,which can realizes fast and accurate classification of malicious software on terminal devices.