An Analysis of Network Protocol Vulnerability Mining Using Fuzz Testing Combined with Deep Learning Models

Abstract

The protection of industrialized management systems and related network protocols is guaranteed by vulnerability mining innovation. The inadequate receiving efficiency and insufficient vulnerability mining capacity of vulnerability mining strategies are their drawbacks. So, this study analyzes the network protocol vulnerability mining using fuzz testing combined with deep learning (DL). In this study, Modbus TCP is employed as a network protocol regarding vulnerability mining. This paper presents a unique threshold-sample-driven deep neural network (T-s DNN) framework. Based on the TSDNN, we construct a fuzzing framework (T-s DNN Fuzzer) for Modbus TCP protocols. The DNN algorithm is first trained to understand the meaning of the protocol's data unit using this framework. The likelihood distribution of every value in the information is quantified using the softmax mechanism. The technique then examines the highest likelihood and the random variable's threshold in deciding whether to use the information value with the optimal likelihood in place of the existing information value. The MBAP header has been finished by the protocol standard. Fuzz tests demonstrate that in addition to increasing sample receipt levels and exploitability, fuzzing devices can identify protocol vulnerabilities rapidly. Experiments conducted with the T-s DNN fuzzer demonstrate that it can detect industrial control protocol vulnerabilities greater in addition to increasing test case reception scores and exploitability.