•  
  •  
 

Abstract

This study proposes an all-weather autonomous foreign-object detection system capable of operating in low-light environments. We used a highly mobile quadcopter equipped with an edge-computing device as the inspection platform. On the software side, we integrated a low-light image enhancement model, a path-tracking algorithm, and a foreign object detection model to achieve rapid object detection. A lightweight, low-light image-enhancement model reduces the need for additional lighting equipment, thereby enabling reliable detection under low-light conditions. The path-tracking algorithm integrates a lightweight semantic segmentation model with traditional image-processing techniques to establish flight paths efficiently and reliably. The foreign-object detection component employs state-of-the-art techniques and is trained on domain-specific datasets to meet specific operational requirements. Upon detecting a foreign object, the system transmits information to the ground station via communication streams, triggering an alert. This technological development enables rapid deployment of an inspection system, addresses the inefficiencies in manual inspections, and significantly enhances operational effectiveness.

Download
COinS