Font Size: a A A

Research And Implementation Of Android Malware Detection System Based On Multi-feature And Attention Mechanism

Posted on:2020-01-16Degree:MasterType:Thesis
Country:ChinaCandidate:Q LuFull Text:PDF
GTID:2428330590954829Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the advent of the intelligent age,people's smart life has gradually deepened from smart phones to smart TVs to smart homes.The application of the Android platform has gradually expanded from mobile phones to various smart devices.As the Android system participates in all aspects of people's lives,its personal information is also processed more and more,and various applications based on Android are growing.However,the malware of the Android platform has also increased.The data shows that in 2018,the number of new virus packages on the Android platform reached 8 millions,and the number of payment-type virus packages increased by 59,000.It can be seen that researching Android malware detection is increasingly important to protect people's personal information and property security.The static methods of Android malicious code detection in the past are mostly based on the single surface feature of APK files.However,with the concealment of malicious code,the detection method of single surface features is not enough to deal with malicious code detection.The rapid development of deep learning has made it a good achievement in the field of text processing and image processing.It is to construct a multi-layer network,combine low-level features through the network and abstract high-level features to represent the object's attribute categories or features.Helps to further classify or predict objects.Apply it to the shallower machine learning method in the Android malicious code detection field,which can better mine the hidden features of malicious code.This paper analyzes the current situation of malware detection at home and abroad,combines the surface features of malware with deep learning to improve the accuracy of detection,and proposes an Android malware detection based on multi-features and attention mechanism.Method,and designed and developed a malware detection system.Through nearly one year of hard work,the following research work has been mainly done:1.Research and analyze the features and models of Android malware detection.The characteristics and models of the current detection of Android malware are studied and analyzed to understand the research status,and the detection effects of various features and the detection effects of various detection models are compared.Through analysis of previous research,understand the analysis process of Android malware,and summarize the techniques required for various detection methods.2.A deep learning model based on attention self-encoding is proposed.The attention mechanism and self-encoder of this paper are introduced in detail from the basic principles,typical applications,advantages and disadvantages.Combined with the traditional features of multi-opcode statistical features,texture fingerprint and combination extracted from this paper,a deep learning model based on attention self-encoding is proposed to extract features twice and obtain more abstract and effective features for classification.Detection.The experimental results show that the method shows good results for malicious code detection.3.Developed the Android malware detection system.By analyzing various feature extraction algorithms and detection model algorithms proposed in this paper,a detection tool based on Android system is designed and implemented.The system puts the computing part on the web server side,and the mobile terminal mainly assumes the display and upload functions.Through testing,the system implements the expected detection function on the Android side.In the research process of this paper,I learned the architecture and security mechanism of Android system,analyzed and extracted various features of Android malware,and proposed a deep learning model based on attention self-encoding to detect Android malicious code,combined with the model.Designed and developed a malicious code detection system based on Android.The research shows that the accuracy of Android malicious code based on the self-encoding deep learning model can reach 89%,which can be effectively applied to malicious code detection,and the multi-feature extraction can improve the accuracy of model detection to 92%.The development of Android-based malware detection system has certain reference significance for mobile security researchers,which is convenient for researchers to extract features and deeply study various detection methods.
Keywords/Search Tags:Android malware, multi-opcode statistical feature, Image feature, Attention machanism, Self-decoder, Detection system
PDF Full Text Request
Related items