Mobile Application Behavior Recognition Based On Deep Learning

Due to privacy protection and other reasons,more and more mobiles should adopt encrypted and secure transmission.Recognition of mobile application behavior through network traffic has become a hot research direction.The research in this paper mainly includes three aspects: the collection of mobile application behavior data sets,the modeling of behavior recognition models,and the modeling of behavior recognition models in small sample scenarios.The specific work is as follows:First,in response to the problem of insufficient data sets,use Appium to automatically operate mobile application behaviors based on mobile phones to realize automatic collection of target application behavior traffic and provide a data basis for subsequent research;Second,in response to the problem of similar traffic in mobile application behaviors,a deep learning method combining dual-domain attention and depth separable convolution was used to construct a behavior recognition model(DSFA),and the original traffic was converted into a grayscale image for direct feature While learning,the dual-domain attention in the model is used to improve the accuracy of behavior recognition,and the deep separable convolution is used to improve the efficiency of behavior recognition.Experiments show that this method is accurate in experiments while reducing the amount of model calculations.Compared with the existing methods,the rate has been greatly improved;Third,for the behavior recognition problem in the case of small samples,a mobile application behavior recognition method based on meta-learning is proposed.The attention-based behavior recognition model of the above research is used as the basic model,and the meta-learning strategy is used to make the recognition model face The new behavior recognition task can be quickly adapted to achieve efficient and accurate recognition of mobile application behavior in the case of small samples.