Utilization of Deep Learning and Unmanned Aerial Vehicle (UAV) for Monitoring Plastic Debris

Plastic debris has become a significant environmental problem due to its persistence in nature and harmful ecological impacts. Recently, Unmanned Aerial Vehicles (UAVs) and deep learning models have been utilized for detection and analysis, as effective plastic debris monitoring is essential for mitigation efforts. While extensive research has focused on coastal areas, riverine environments remain understudied. This study evaluates the capability of UAVs and the You Only Look Once version 5 (YOLOv5) deep learning model in detecting and quantifying macro plastic debris in river systems. The research was conducted in 2021 in the lower reaches of the Citarum River, Muara Gembong District, Bekasi Regency, West Java Province, Indonesia. UAVs were deployed to capture aerial images, which were then processed using the YOLOv5_small and YOLOv5_large models. A total of 416 images were analyzed, and plastic debris was classified based on size and distribution. The results indicate that UAV effectively capture high-resolution images of plastic debris in river environments. The YOLOv5 model successfully identified plastic debris; however, its accuracy requires further improvement, with mean Average Precision (mAP) scores of 16.90% and 19.21% for YOLOv5_small and YOLOv5_large, respectively. Detection performance varied depending on factors such as solar radiation intensity, surface water reflection, and debris characteristics. Future research should explore advanced deep learning architectures, integrate multi-spectral imaging, and assess seasonal variations in plastic debris accumulation. This study provides a foundation for the development of automated, scalable plastic debris monitoring systems to support environmental management and pollution mitigation efforts.