AI‐Driven Secure Map Construction for UAV Inspection in Edge‐Cloud Collaborative IIoT Systems

ABSTRACT In response to the challenges faced by UAV inspection systems in dynamic industrial IoT environments, such as adversarial attacks, real‐time mapping accuracy, and computational resource limitations, we propose a security‐enhanced visual SLAM framework integrating AI and edge‐cloud collaboration. The framework achieves robust environmental perception and tamper‐resistant map construction in complex scenarios through multisensor data fusion and lightweight adversarial defense mechanisms. Key innovations include: a spatial–temporal consistency‐based target segmentation architecture using YOLOv7 and Mask R‐CNN, effectively suppressing dynamic interference and adversarial disturbances; an adaptive ORB‐SLAM3 optimization scheme for resource‐constrained UAVs, combining inertial navigation and semantic feature filtering to enhance localization stability; and a privacy‐preserving edge‐cloud task scheduling strategy based on federated learning and threshold cryptography, ensuring data integrity in distributed computing. Experimental results demonstrate that the proposed system significantly improves environmental modeling accuracy and security capabilities in dynamic industrial settings, providing a technical paradigm for intelligent monitoring of critical infrastructure that balances real‐time performance, security, and scalability.