一种Android恶意软件检测模型  

Android malware detection model

作　　者：杨宏宇[1] 那玉琢 YANG Hongyu;NA Yuzhuo(School of Computer Science and Technology,Civil Aviation Univ. of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院

出　　处：《西安电子科技大学学报》2019年第3期45-51,共7页Journal of Xidian University

基　　金：国家自然科学基金民航联合研究基金(U1833107);国家科技重大专项(2012ZX03002002)

摘　　要：针对传统Android恶意软件检测方法检测精度较低等不足,提出一种基于双通道卷积神经网络的Android恶意软件检测模型。首先,提取应用程序的原始操作码序列并生成指令功能序列;然后,将两种序列分别作为卷积神经网络两个通道的输入迭代训练并调整各层神经元权重;最后,通过已训练的检测模型实现对Android恶意软件的检测。实验结果表明,该检测模型对恶意软件具有较好的检测分类精度和检测准确率。Aiming at the low detection accuracy of traditional Android malware detection technology, an Android malware detection model based on the Dual-channel Convolutional Neural Network (DCNN) is proposed. First, it extracts the software original opcode sequence and generates the command function sequence. Then, it uses these two sequences as the input to the two channels of the convolutional neural network to iteratively train and adjust the neurons weights in each layer. Finally, the trained detection model implements the Android malware detection. Experimental results demonstrate that our detection model has a good detection accuracy and detection precision for malware.

关 键 词：恶意软件 分类检测 操作码序列 指令功能序列 卷积神经网络 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]