WEB Application Attack Detection Based On Deep Learning

Web application attack detection is an important task in network security,which aims to detect whether the request sent by the user to the server is an attack request through various methods.With the increasing number of web applications and the complexity of data,the accuracy and stability of web application attack detection become particularly important.Web application data is more complex than image,audio and other data.Therefore,using the existing deep learning model directly can not achieve better detection results.Therefore,this paper starts with the existing deep learning model and improves it to make it better suitable for the field of Web application attack detection.This paper mainly studies two methods of classification detection and anomaly detection.Classification detection uses normal and attack data to train the model,which has a high detection accuracy.Aiming at the disadvantage that the existing deep learning model is easy to be affected by the input length,a dynamic feature LSTM(long short term memory)model is proposed in this paper to improve the accuracy and stability of detection.Anomaly detection can use unlabeled data for training,avoiding the shortcomings of data tagging.In this paper,we mainly study the anomaly detection method based on reconstruction error,and improve the detection method using autoencoder as a reconstruction model,and propose the detection method based on the attention gated convolution network reconstruction model,which improves the detection accuracy of the model.The main work of this paper is as follows:(1)The existing web application attack detection methods are studied.(2)The method of web application attack classification detection based on deep learning is studied.Among them,we mainly study the detection method based on textcnn(text revolutionary neural network)and LSTM as the classification model.Because of the shortcomings that textcnn and LSTM are easy to be affected by input length as classification models,this paper proposes a dynamic feature LSTM detection model,which can dynamically extract features related to classification tasks.Experimental results show that the model not only has high detection accuracy but also has strong stability,which is not easily affected by the input length.(3)The anomaly detection method of web application attack based on deep learning is studied.Firstly,the basic framework of anomaly detection method based on reconstruction error is studied.In this framework,the detection method based on self encoder is studied.In view of the shortcomings of this method,this paper proposes a detection method based on the self attention gate convolution network as the reconstruction model.The experimental results show that the detection accuracy of this method is higher than that of the self encoder based detection method.(4)Summarize and refine the contents of the paper,and analyze the problems still existing in the paper.And analyze future work.