文献类型：期刊文献

英文题名：Comparative assessment of manual and UAV-based deep learning approaches for beach litter

作者：Li, Haiwei[1];Jiang, Shiqi[2];Xiong, Zhengye[1];Dai, Zhenqing[3];Sun, Ruikun[3];Li, Chengyong[2]

机构：[1]Guangdong Ocean Univ, Sch Elect & Informat Engn, Zhanjiang 524088, Peoples R China;[2]Guangdong Ocean Univ, Analyzing & Testing Ctr Ocean Western Guangdong Pr, Coastal Ecol Engn Technol Res Ctr Zhanjiang, Analy & Testing Ctr, Zhanjiang 524088, Peoples R China;[3]Guangdong Ocean Univ, Sch Chem & Environm, Zhanjiang 524088, Peoples R China

年份：2026

卷号：540

外文期刊名：JOURNAL OF CLEANER PRODUCTION

收录：SCI-EXPANDED(收录号：WOS:001664209100001)、、EI(收录号：20260419953683)、Scopus(收录号：2-s2.0-105027934369)、WOS

基金：This work was supported by the National Natural Science Foundation of China (22474027) , Innovation Team Project of Universities in Guangdong Province (2025KCXTD020) , and Scientific Research Start-up Funds of Guangdong Ocean University (060302122402) . We would also like to thank the Analytical and Testing Center of Guangdong Ocean University, the Guangdong Provincial Observation and Research Station for Tropical Ocean Environment in Western Coastal Water, and the Guangdong Provincial Key Laboratory of Intelligent Equipment for South China Sea Marine Ranching for providing facilities, technical support, and valuable assistance throughout this study. We sincerely thank the editor and the anonymous reviewers for their constructive comments and insightful suggestions, which greatly improved the quality of this manuscript.

语种：英文

外文关键词：Marine litter; Plastic waste; Manual monitoring; UAV monitoring; Deep learning; YOLO model; China

外文摘要：Since the 1980s, marine litter has increasingly accumulated on global coastlines. Traditional monitoring methods are time-consuming, labor-intensive, and face significant challenges in consistently detecting small or partially buried litter, which has driven the need for intelligent, data-driven approaches. This study proposes an optimized UAV-deep learning framework to address these challenges. We selected two typical coastal regions in Guangdong Province, China: Donghai Island (Zhanjiang) and Daya Bay (Huizhou). At these sites, a high-resolution dataset consisting of 6300 UAV images through aerial photography and manual annotation. An enhanced YOLOv8 model with Gold-YOLO-inspired feature extraction and fusion was developed and trained on this dataset. The optimized model achieved 96.50 % precision, 87.00 % recall, 91.50 % F1-score, and 89.80 % mAP@0.5, markedly outperforming baseline YOLO variants and traditional manual methods in both accuracy and efficiency. Field validation at 12 sampling sites showed that the system improves operational efficiency by 75 %, completing surveys in hours instead of days. Environmental assessment using the Clean Coast Index (CCI), Plastic Abundance Index (PAI), and Hazardous Litter Index (HLI) revealed that plastics account for 85 % of all detected litter. The proposed UAV-deep learning framework enables rapid, accurate, and cost-efficient large-scale beach litter monitoring, providing coastal managers with a practical tool for data-driven environmental management and pollution control. It paves the way for integration with automated cleanup systems and long-term ecological risk assessment, contributing to global marine plastic pollution mitigation.