Surveillance pour la détection et la localisation des cyberattaques dans les réseaux sans fil

In an era of pervasive wireless connectivity, billions of smart devices and sensors seamlessly interlink to form the Internet of Things (IoT). This ubiquity of wireless technology brings immense benefits but also exposes new cyber-physical vulnerabilities. This dissertation investigates advanced monitoring techniques to detect and localize cyber attacks in wireless networks, addressing threats ranging from rogue drones breaching physical perimeters to impostor devices mimicking legitimate network nodes. We pursue a two-pronged approach. Part I of the thesis focuses on physical intrusion detection: we develop an audio-visual drone detection framework that fuses acoustic signatures and infrared imagery via deep learning to overcome the limitations of single-sensor systems, and we introduce TRIDENT, a tri-modal intrusion detection engine that combines audio, video, and Radio Frequency (RF) signals for real-time detection of unauthorized aerial targets. Part II concentrates on physical-layer device authentication: we design a machine learning-assisted hybrid physical-layer authentication scheme that leverages inherent radio hardware impairments (e.g., oscillator offsets) to robustly verify wireless devices in multi-node networks, and we propose a multi-band RF fingerprinting technique capable of distinguishing myriad IoT devices—like finding a needle in the spectrum haystack—even under dynamic channel conditions. Through comprehensive experiments and evaluations, our contributions demonstrate enhanced detection accuracy, rapid response times, and improved resilience against adversarial conditions. Collectively, these results advance the state of the art in IoT security, providing layered defenses that augment both the physical security of the environment and the trustworthiness of wireless communications.