Faster R-CNN With Variable-Sensitive Map for UAVs Fault Detection: Addressing Simultaneous Sensor Faults in Autonomous Flights

Unmanned aerial vehicles (UAVs) have advanced significantly and are increasingly used in civil and military field. Ensuring their safety and reliability is crucial, and fault detection is central to secure autonomous flight. Currently, data-driven fault detection methods have gained significant attention due to their advantage of not relying on aircrafts’ dynamics. However, existing researches rarely address the simultaneous fault detection problem for autonomous UAVs during flight. To tackle this challenge, this paper proposes a Faster R-CNN model with variable-sensitive map (VSM). First, data imagefication and image enhancement technologies are employed to facilitate its better adaptation to this model in machine vision realm. Then, a novel multiple feature map prediction (MFMP) based on VSM is proposed and applied to Faster R-CNN. Furthermore, by modifying the labeling strategy, the model’s accuracy in identifying simultaneous faults is improved. The experimental results show that models in the field of machine vision are feasible for fault detection in autonomous UAVs and achieve high accuracy compared to mainstream data-driven methods. Additionally, experiments designed to detect simultaneous faults demonstrate that the proposed method performs better.