Research On SQL Injection Detection Based On Deep Learning

With the advent of the Web 2.0 era,network technology has developed rapidly,and at the same time,network security issues have become increasingly prominent.In OWASP's list of vulnerabilities for ten consecutive years,injection vulnerabilities ranked first,and SQL injection vulnerabilities are the most common manifestation of injection vulnerabilities.This vulnerability may lead to information and data leakage,which in turn will cause huge losses,and due to its complex form of vulnerabilities and variable attack methods,the attack and defense around SQL injection has become an important research topic in network security.Many security researchers and security Manufacturers invest a lot of time and energy in research.For the detection of SQL injection,mature commercial products at home and abroad are mainly carried out through log traffic analysis and combined with rule matching,but with the upgrade of offense and defense,they have been unable to meet the needs of network security.With the vigorous development of machine learning,the security industry has also begun to conduct intrusion detection research in conjunction with related technologies.Intrusion detection is achieved by converting traffic data into text information and then passing it to a learning model for analysis and processing.However,the detection results are uneven,which is mainly caused by the following two aspects: the lack of data set confusion,and imperfect feature vector extraction.Based on the above analysis,this article carries out research work in turn from the following three perspectives:First,this article conduct web security related research,focusing on the causes and attack methods of SQL injection vulnerabilities.The secondary development and utilization of the open source tool SQLMAP,combined with a more effective sample generation method,generates higher quality SQL injection sample data.After that,combined with the actual WAF and IDS bypass technologies,more efficient variant samples are generated,which improves the confusion and quality of the sample data.Secondly,the research incorporates samples of diversity variants.Convolutional neural networks with different dimensions are used for detection and comparison,and different training sets are used for comparison to obtain a more effective detection model.Finally,a SQL injection intelligent detection model is designed and implemented based on the convolutional neural network algorithm.And through multiple iterations of different model parameters,model optimization is achieved.It greatly solves the problem of low validity of data set samples.At the same time,the comparison of different dimensions improves the detection accuracy of SQL injection attacks and effectively reduces the problem of overfitting the model.This paper conducts experimental demonstrations from different dimensions of convolutional neural networks and out-of-order data sets: through the injection of SQLMAP after the secondary development of SQLMAP and the higher availability and higher availability Verify the advantages of the detection model in this paper.Through the differences between different black and white sample sets,the performance and differences of the deep learning model are compared to verify the advantages of the detection model in this paper,thereby verifying the rationality and usability of the detection.