具有鲁棒性的安卓恶意软件检测模型研究

he purpose of this research is to realize an Android malware adversarial detection model that can resist attacks. The research includes Android function call graphs, graph neural networks (GNN), and adversarial detection. We propose an Android malware detection model based on graph convolutional neural network (GCN), which uses the function call graph as the graph structured data, uses the graph neural network as the classifier, and design a method to reduce the scale of the function call graph. For some common attack methods, we also designed corresponding defense metResearch of Robust Android Malware Detection Modelhods, which greatly enhanced the robustness of the model. Experiments show that our method can effectively detect 97.5% of Android malware, this number also reached 93.3% under adversarial attack. Our experiments on real datasets not only show the effectiveness of the method proposed in this article, but also show the effectiveness of the graph neural network model in Android malware detection and adversarial detection.)