Research On Malicious Web Page Recognition Based On Image Semantic Understanding

Information security is closely related to users' property and personal interests.In order to prevent users from mistakenly entering malicious web pages and suffering property losses,how to identify the actual content of web pages efficiently and efficiently has become a hot topic of current research.However,with the rapid development of today's information industry,the rate of information transmission has been greatly improved.Malicious web pages have gradually transited from text as the main information carrier to the mode of combining image and text.Especially in gambling,pornography,fraud and other webpages,there are a lot of harmful picture information.If these image features can be extracted effectively,it will help to improve the recognition rate of malicious web pages.Therefore,this paper improves the recognition effect of malicious web pages by extracting the feature semantics of web images and combining the text features of web pages.The main work of this paper is as follows:(1)In order to effectively extract image information from malicious web pages,this paper uses in-depth learning method to recognize images,aiming at extracting image information to assist malicious web page detection.By summing up the advantages and disadvantages of common deep learning algorithms,design comparative experiments,and select the algorithm based on the actual detection environment of malicious web pages.At the same time,in order to improve the network structure and detection efficiency of MASK R-CNN,there is computational redundancy in MASK R-CNN.(2)Because of the small amount of information that can be expressed in static images,especially the ambiguity of attitude information due to the lack of detailed expression,this paper uses the method of combining semantic segmentation with Kinect template matching to predict the semantics of the target attitude.By recognizing the basic human body posture,and combining the object context information in the image,the BP neural network is used for semantic combination.And deduce the final semantics.(3)Study the attack principle and detection methods of malicious web pages.Aiming at the inadequate use of image information in existing malicious web page detection technology,improve the traditional malicious web page detection algorithm,introduce image semantic extraction module to assist web page recognition,and then fuse with other features of malicious web pages,and finally send it to classifier for recognition.In this paper,the malicious web page detection system is implemented by modular design.At the same time,different types of web page data sets are used to test the system to verify the correctness of the improved algorithm.The test results show that,compared with traditional malicious web page detection methods,the detection accuracy is enhanced via the help ofthe introduction of image semantic features.