Deep Learning-Based Face Fatigue Monitoring Method from the Perspective of Cyber Security

In the context of digital transformation, the deep integration of cyberspace and the physical world has given rise to new types of security challenges, and there is an urgent need for real-time and reliable personnel identity verification and behavioural status monitoring in fields such as intelligent transportation and telemedicine. As a typical human-caused security risk, fatigued driving not only threatens road traffic safety but also may be used by cyberattackers to implement identity fraud or data theft, resulting in derivative hazards. To this end, this study proposes an improved YOLOv6 facial fatigue monitoring method that integrates cybersecurity perspectives; combines traditional computer vision techniques with cybersecurity modelling; is based on the PyTorch framework; integrates OpenCV, Dlib, and multilevel fatigue assessment algorithms; optimises the PP-LCNet lightweighting module and the ShuffleNetv2 channel rearranging mechanism by incorporating the YOLOv6 backbone network; reconfigures the neck feature fusion layer to enhance the multiscale feature extraction capability; and introduces the adversarial sample training mechanism to improve the model’s defense capability against network attacks. Experiments show that the improved model significantly improves the fatigue feature recognition accuracy and real-time performance on the embedded platform, which provides innovative technical support for the construction of networked human security monitoring and intelligent traffic active defense systems.