Android Malicious App Detection Based on CNN Deep Learning Algorithm
Peng Ju He, Gang Gan · IOP Conference Series Earth and Environmental Science · 2020
Abstract In recent years, with the increasing number of Android applications, Android malicious apps have also increased exponentially. This paper proposes a detection scheme based on CNN deep learning algorithms. The scheme uses static analysis and detection methods. CNN deep learning algorithms have become very mature in the field of image classification. Therefore, this article uses a conversion algorithm to convert Android APK binary files to RGB PNG images, and then uses the algorithm to extract permissions in the APK, and converts the permission matrix to a transparency matrix, combines the previous RGB images, and finally generates RGBA images. RGBA images contain more information than RGB images, and the use of CNN deep learning algorithms works better and the final detection rate reaches 93.4%.