Blended Wing UAV for Intelligent Surveillance: Design and Deep Learning-Based Human Detection

In an era where unmanned aerial vehicles (UAVs) are transitioning from passive observers to intelligent mission partners, integrating aerodynamic efficiency with onboard artificial intelligence (AI) unlocks new capabilities. This paper presents a novel micro-class Blended Wing Body (BWB) UAV with a tail, co-designed through a synergistic optimization of aerodynamic, structural, and computational intelligence parameters. Unlike conventional designs, the aerodynamic configuration and embedded AI model were developed concurrently, ensuring mutual optimization under weight, stability, and processing constraints. The UAV, realized through iterative CAD modeling, CFD simulations, FEM-based structural analysis, and composite fabrication, achieved a lift-to-drag ratio of 17.7 and a structural safety factor exceeding 1.8. The onboard EfficientNetB3-based deep learning model demonstrated 96.3% training accuracy, 94.7% validation accuracy, precision of 0.93, and recall of 0.91. Field trials verified reliable human detection beyond 120 m, with accurate pose classifications such as Shooting-Crouching at 0.8759 confidence. The co-optimization of aerodynamic design and embedded AI inference establishes a new framework for developing lightweight, mission-ready UAVs for surveillance, border security, and rapid payload delivery applications.