Research On Adversarial Example Technology For Io T Device Identification

The security situation of Internet-of-Things(Io T)devices is severe.By identifying devices and obtaining device information,device managers can take targeted security measures.Device attackers can also take targeted attack measures based on the device information.This thesis focuses on the adversarial example technology for Io T device identification,and aims to disrupt the attacker’s identification of the device by generating adversarial examples,thereby protecting the device information from malicious use.There are differences in the application layer packets of Io T devices,which often carry relevant information of the device.However,the existing device identification technology ignores the relationship between device information,cannot identify devices in a fine-grained or high-precision manner,and lacks the ability to identify unknown devices,thus cannot provide reliable and powerful adversarial targets for adversarial examples.At the same time,there is a lack of research on adversarial example technology for Io T device identification,and the application-layer packets used for identification is different from images and natural language text.To address the above problems,the main research points and contributions of this thesis are as follows:(1)This thesis proposes an Io T device identification method based on the entity relationship extraction model,which can extract device information triples(device type,vendor,product)from the application layer packets of Io T devices.This method takes the device type,vendor,and product model as the subject and object in the relationship through the defined relationship type,and jointly extracts the entity and relationship of device information.In addition,the method identifies entities,the subjects and objects of relationships in the packets based on word pairs,which solves the problem of multi-word entities and overlapping relationships.This method makes full use of the semantic information of Io T device packets and the potential relationship between device information,realizes fine-grained identification of Io T devices,and improves the ability to identify unknown devices.The thesis conducts experiments to verify the effectiveness of Io T device identification.The thesis evaluates the identification effect of known devices on more than 10,000 test packets,and the F1 score of the model can reach 90.99%;evaluates the identification effect of unknown devices on 1225 packets,and the F1 score of the model reaches 79.10% in the dataset that contains recognition results.(2)This thesis designs an adversarial example generation technique for Io T device fingerprints.Considering that the fingerprint model of the attacker is a black box in real scenarios,this thesis transforms the generation of adversarial example into a combinatorial optimization problem.The adversarial example of Io T device fingerprint should not only affect the threat model of hackers,but also cannot affect the normal use of ordinary users.This thesis proposes that the adversarial examples need to meet the requirements of structural,visual and semantic consistency,and designs a search space construction method that combines region division,character perturbation and word-level perturbation.In addition,the thesis designs two search algorithms to find adversarial examples in the search space.This method can effectively interfere with the device identification of hackers or attackers in a black-box scenario,while ensuring the use and access of ordinary users.The experiments show the method of generating adversarial examples of Io T device fingerprints in this thesis can achieve a success rate of 85.60% in perturbing the product model among 1000 packets,and the generated adversarial examples are difficult for humans to perceive.