An Object-Oriented Nighttime Light Classification Based on Light Color Temperature: A New Perspective From AAV Nighttime Images

The nighttime urban environment is increasingly affected by various forms of artificial light at night. Color temperature, a critical characteristic of light, has significant effects on numerous fields and industries. The widespread adoption of light-emitting diode (LED) light, a low-carbon technology, has resulted in extensive use of lights with varying color temperatures in diverse settings. However, it is crucial to recognize that different color temperatures have distinct impacts on human health and ecological systems. Therefore, understanding the spatial distribution and composition of nighttime light (NTL) with different color temperatures is essential for developing sustainable strategies that balance public safety, energy consumption, and ecosystem conservation. In response to this need, we propose a color temperature-based lighting source classification system utilizing autonomous aerial vehicle (AAV)-captured NTL images, rather than the traditional satellite-based NTL images, due to the superiority of spatial resolution (SR). We employ an object-oriented classification method to categorize lights into high-pressure sodium (HPS), warm LEDs, cool LEDs, and colored LEDs. Moreover, to evaluate the effect of flight altitude on classification accuracy, we classify lights at seven different altitudes and compare their accuracy at each level. Our results indicate that the random forest (RF) algorithm can accurately identify the four types of lights, with the highest classification accuracy achieved at a flight altitude of 350 m, where the overall accuracy (OA) and kappa coefficient were 0.957 and 0.947, respectively. Moreover, at this altitude, the highest producer’s accuracy (PA) for warm LEDs and colored LEDs was 0.971 and 0.942, respectively, while the user’s accuracy (UA) for each light type exceeded 0.9. In addition, the methodology also demonstrated strong performance in more complicated regions, as evidenced by an off-site application accuracy of 0.847 and a kappa coefficient of 0.808. This study is the first to identify NTL types based on color temperature, offering a new perspective for urban lighting planning and light pollution management.