基于无人机遥感与机器学习的岸滩大型塑料垃圾监测方法

鉴于现场人工调查手段对海滩垃圾的监测具有成本高、耗时长等特点，近年来无人机遥感技术在海洋垃圾监测中的应用受到广泛关注。为评估无人机遥感及机器学习技术在海洋垃圾信息收集与分析中的有效性，首先，本研究将无人机遥感图像的目视解译结果与实地调查进行对比，结果表明，目视解译对垃圾的总体识别率达68.1%～75.1%，对各类塑料垃圾的识别率达38.9%～94.3%，说明无人机图像解译垃圾具有较强的可行性；其次，在目视解译无人机图像的基础上，引入机器学习技术自动解译图像，该技术对各类塑料垃圾的识别准确率达77.9%～81.5%，回收率达45.6%～60.6%；最后，本研究将无人机遥感与机器学习技术运用于崇明岛岸滩的实地调查中，成功地实现了样区塑料瓶、塑料渔网、塑料浮球、塑料泡沫和浮木等多类垃圾通量和组成的快速统计与空间分布分析，并揭示了台风前、后岸滩垃圾的变化特征。本研究发展并验证了基于无人机遥感与机器学习图像解译的岸滩塑料垃圾监测技术，实现了垃圾数据的半自动化采集、识别与统计的完整监测过程，提升了监测效率。本研究发展的机器学习模型可解决海量、复杂无人机影像中垃圾的定位与识别需求，一定程度上实现了智...

Monitoring of marine macro-plastic litter in the coastal zone based on UAV remote sensing and machine learning

Due to the high cost and low efficiency of field investigation, unmanned aerial vehicle (UAV) remote sensing has been applied to marine litter monitoring worldwide in recent years. To evaluate the effectiveness of UAV remote sensing and machine learning in marine litter monitoring, firstly, we compared visual interpretation result with those obtained from field investigation. The results showed that the overall recognition precision of marine litter reached 68.1%~75.1%, while the recognition precision of all kinds of plastic litter was 38.9%~94.3%, proving the reliability of image interpretation. Secondly, based on the visual interpretation of litter remote sensing images, machine learning method was introduced to image interpretation. The recognition precision of various types of plastic litter reached 77.9%~81.5%, and the recovery rate was 45.6%~60.6%. Finally, this method was successfully applied to analyze the flux, composition and spatial distribution of plastic litter in Chongming Island and to reveal the differences between before and after the typhoon. In brief, we developed and verified the application value of UAV remote sensing and machine learning to detect marine macro-plastic litter. We realized the complete monitoring process of semi-automatic data collection, identification and statistics, which improved the monitoring efficiency. Machine learning model developed in this paper can be used to solve the demand of litter location and identification in massive UAV images. This intelligent monitoring and evaluation method can support further scientific research and management of marine macro-plastic litter in coastal zone.