Detecting command injection attacks in web applications based on novel deep learning methods

With the continuous advancement of science and technology, while the use of the Internet has brought great convenience to people, it has also aggravated the emergence of network security problems, especially web application security. In the field of web application security, web command injection attacks pose an important security threat, with extremely high levels of harm. Attackers can cause server information leakage or even severe server paralysis by executing relevant malicious commands. As the malicious confusion and number of web application attacks gradually increase, traditional web command injection detection methods have gradually exposed many flaws. These include the model’s feature extraction process being too complex, the model’s poor recognition of malicious code, low recognition efficiency, too high a false positive rate, etc. Under the trend of increasingly serious web security problems, the emergence of artificial intelligence technology has greatly solved network security problems. Therefore, in response to the above problems, we use deep learning technology to propose a new web command injection attack detection model. By combining the relevant features of web command injection attacks, dual CNN convolution channels are used for hybrid feature extraction, the BILSTM network is used to bidirectionally identify the extracted sentence sequence features, and the attention mechanism is combined with the weight distribution of keyword features. We used our model to train and test on two data sets respectively to verify the effectiveness of our proposed feature extraction method and attention mechanism. Experimental results show that our proposed detection method achieves a precision rate of 99.3% and a recall rate of 98.2% in the actual collected dataset. We tested our model on the public HTTP CSIC 2010 dataset, and the experimental results achieved an accuracy of around 99%. Compared with other traditional detections, our proposed model can identify web command injection attacks more effectively.