Android malware detection method based on new gray image and VGG

Abstract

Android malware is transformed into gray image and recognized by deep learning model, so as to realize end-to-end detection. This method has the natural advantage of resisting code obfuscation attack. At present, most of the methods used by researchers for gray image recognition of Android malware are roughly processed by a single convolutional neural network, which does not make good use of this advantage. Therefore, this paper proposes a new Android malware detection method, which is different from the traditional detection method XML file and class.dex file is purified, combined and mapped into a gray image, and then the VGG neural network model is used to detect the gray image. At the same time, we take out the models such as mobilenet and alexnet for comparison. The experimental results show that the purification method has good Android malware detection effect, and the VGG model training results have higher detection accuracy than the other two algorithms, and all evaluation indexes have good performance.