Research On Mobile Application Security Detection Method Based On Intermediate Represent

With the popularity of smart terminal devices,Android malicious applications also will be fast and sustained growth.In order to deal with the security risk of Android smart terminal,this paper proposes a static detection method based on deep learning theory.The model adopts the permission and Jimple instruction code as the feature,and adopts a variety of features coordination.First,by decompiling the APK file to get the manifest file,read the permission,as a result of the feature.Then use Soot to decompile dex file to obtain the representation of the intermediate language,and then simplify the intermediate language code to obtain the sequence of code.The improved double pooling convolution neural network is used to extract the code sequence,and the characteristic result is obtained.The feature results obtained in the above steps are normalized and trained using a support vector machine to obtain the final result.Experiments show that the use of this method to detect,with a high detection rate.The main work and achievements of this paper are as follows:1)The status and analysis of Android system malicious application,the development and application of the deep learning algorithms and anti-tools2)Apply the Jimple intermediate language to Android malware detection.using Soot tool decompile dex file into Jimple language,analyze the characteristics of its language,reducing the feature dimension.3)optimized the model of the convolution neural network and proposed a double pooling convolution neural network.According to the characteristics of Jimple sequence feature sparse,a double pooling convolution neural network is proposed to accelerate the extraction of instruction sequence features.4)based on the two kinds of features extracted from above steps,put forward the Android static detection model.The use of double pooling convolution neural network to extract the malicious features of the sensitive code segment,and the combination of permission features,constitute a static code for the static characteristics,and support vector machine for classification detection.At the same time,the training method of static detection model is put forward.Experiment and give the results.