Marine litter is a global problem that requires soon management and design of mitigation strategies. Marine litter monitoring is an essential step to assess the abundances, distributions, sinks and hotspots of pollution as well as the effectiveness of mitigation measures. However, these need to be time and cost-efficient, fit for purpose and context, as well as provide a standardized methodology suitable for comparison among surveys. In Europe, the Marine Strategy Framework Directive (MSFD) provides a structure for the effective implementation of long-term monitoring. For beaches, the well-established 100 m OSPAR macrolitter monitoring exists. However, this method requires a high staff effort and suffers from a high spatio-temporal variability of the results. In this study, we test the potential of aerial drones or Unmanned Aerial Vehicles (UAVs) together with a Geographic Information System approach for semi-automatic classification of meso- (1–25 mm) and macrolitter (>25 mm) at four beaches of the southern Baltic Sea. Visual screening of drone images in recovery experiments (50 m2 areas) at 10 m height revealed an accuracy of 99%. The total accuracy of classification using object-based classification was 45–90% for the classification with four classes and 50–66% for the classification with six classes, depending on the algorithm and flight height used. On 100 m beach monitoring transects the accuracy was between 39–74% (4 classes) and 25–74% (6 classes), with very low kappa values, indicating that the GIS classification method cannot be regarded as a reliable method for the detection of litter in the Southern Baltic. In terms of cost-efficiency, the drone method showed high reproducibility and moderate accuracy, with much lower flexibility and quality of data than a comparable spatial-OSPAR method. Consequently, our results suggest that drone based monitoring cannot be recommended as a replacement or complement existing methods in southern Baltic beaches. However, drone monitoring could be useful at other sites and other methods for image analysis should be tested to explore this tool for fast-screening of non-accessible sites, fragile ecosystems, floating litter or heavily polluted beaches.

中文翻译：

---

用于波罗的海海滩大型垃圾监测的空中无人机的效率

海洋垃圾是一个全球性问题，需要尽快管理和设计缓解策略。海洋垃圾监测是评估污染的丰度、分布、汇和热点以及缓解措施的有效性的重要步骤。然而，这些需要时间和成本效益高，适合目的和背景，并提供适用于调查之间比较的标准化方法。在欧洲，海洋战略框架指令 (MSFD) 为有效实施长期监测提供了一种结构。对于海滩，存在完善的 100 m OSPAR 大型垃圾监测系统。然而，这种方法需要大量的工作人员努力，并且结果的时空变化很大。在这项研究中，我们测试了空中无人机或无人驾驶飞行器 (UAV) 以及地理信息系统方法的潜力，用于在波罗的海南部的四个海滩对中型（1-25 毫米）和大型垃圾（> 25 毫米）进行半自动分类. 在 10 m 高的恢复实验（50 m2 区域）中对无人机图像进行目视筛选，显示准确率为 99%。根据所使用的算法和飞行高度，基于对象的分类的总准确率对于四类分类为 45-90%，对于六类分类为 50-66%。在 100 m 海滩监测断面上，准确度在 39-74%（4 级）和 25-74%（6 级）之间，kappa 值非常低，表明 GIS 分类方法不能被视为检测波罗的海南部垃圾的可靠方法。在成本效益方面，无人机方法显示出高重现性和中等准确性，与可比的空间 OSPAR 方法相比，灵活性和数据质量要低得多。因此，我们的结果表明，不能推荐基于无人机的监测作为波罗的海南部海滩现有方法的替代或补充。然而，无人机监测在其他地点可能很有用，应该测试其他图像分析方法，以探索这种工具，以快速筛选无法进入的地点、脆弱的生态系统、漂浮的垃圾或污染严重的海滩。与可比较的空间 OSPAR 方法相比，灵活性和数据质量要低得多。因此，我们的结果表明，不能推荐基于无人机的监测作为波罗的海南部海滩现有方法的替代或补充。然而，无人机监测在其他地点可能很有用，应该测试其他图像分析方法，以探索这种工具，以快速筛选无法进入的地点、脆弱的生态系统、漂浮的垃圾或污染严重的海滩。与可比较的空间 OSPAR 方法相比，灵活性和数据质量要低得多。因此，我们的结果表明，不能推荐基于无人机的监测作为波罗的海南部海滩现有方法的替代或补充。然而，无人机监测在其他地点可能很有用，应该测试其他图像分析方法，以探索这种工具，以快速筛选无法进入的地点、脆弱的生态系统、漂浮的垃圾或污染严重的海滩。