Assessing nighttime artificial light pollution from the perspective of an unmanned aerial vehicle tilt

Abstract

Increasing artificial light at night (ALAN) impacts urban sustainability and contributes to light pollution. Nighttime satellites miss side ALAN, so drone-captured tilted images and measured illuminance are used to assess ALAN pollution within urban streets. By integrating deep learning methods, ALAN information is efficiently extracted and statistically analyzed. Key findings include: (1) With the drone camera set to f2.8, the best correlation between image RGB color channel brightness and measured illuminance occurs at ISO 200 and 1/4s, achieving R2 reaches 0.83 and MAE reaches 7.89 lx. (2) Mask R-CNN and Deeplabv3 achieve over 0.91 in extraction accuracy, with Mask R-CNN excelling in window, street, and neon lights, while Deeplabv3 better handles building sides and road surfaces. (3) Vertical light pollution from typical residential buildings shows a gradual attenuation trend with increasing height. This research presents a novel technical approach for monitoring urban ALAN pollution through UAV-captured tilted images.

Keywords

• UAV tilted images
• artificial light at night
• light pollution
• illuminance
• deep learning