耦合SVM和Cloud-Score算法的Sentinel-2影像云检测模型研究

云覆盖阻碍了光学遥感卫星对地观测的有效范围，快速、准确的云检测是遥感应用产品生成过程中的重要一步。针对Google Earth Engine云平台中缺乏适用且高质量的云检测模型，以热带多云的斯里兰卡为研究区，构建了耦合SVM和Cloud-Score算法的Sentinel-2影像云检测模型，通过实验从目视判读与定量分析两个角度对比了其与QA60法、Cloud-Score算法以及Fmask的云检测精度，并在海南岛和亚马逊森林两个地区进行了云检测测试。研究结果表明：Fmask模型的云检测性能最低，总体精度仅为63.45%，存在严重的水体误分为云的现象，但其漏提率极低；QA60法对卷云识别不足，漏提率较高，同时存在一定的误分现象，并且低空间分辨率影响了云体边界提取结果的细节性；Cloud-Score算法的云检测性能明显好于QA60法，总体精度达到了89.83%，误提率仅为2.17%，但仍存在部分卷云漏提的现象；相比于其他3种云检测方法，本文提出的云检测模型总体精度最高，达到了98.21%，并且拥有极低的漏提率和误提率，能比较精准地识别出云体的边界，可满足Sentinel-2遥感产品的云检测预处理需求。

Research on Cloud Detection Model of Sentinel-2 Image Coupled with SVM and Cloud-Score Algorithm

Cloud coverage hinders the effective range of earth observation by optical remote sensing satellite. Rapid and accurate cloud detection is an important step in the product generation process of remote sensing application. In view of the lack of suitable and high-quality cloud detection model in Google Earth Engine cloud platform， this study takes tropical cloudy Sri Lanka as the study area， constructs a Sentinel-2 image cloud detection model coupled with SVM and Cloud-Score algorithm. Through experiments， the cloud detection accuracy of this method is compared with that of QA60 method， Cloud-Score algorithm and Fmask from the point of view of visual interpretation and quantitative analysis. The results show that the cloud detection performance of Fmask model is the lowest， and the overall accuracy is only 63.45%. It has a serious phenomenon that water body is mistakenly divided into clouds， but its omission error is very low. The QA60 method has insufficient recognition of cirrus clouds， and the omission error is high. At the same time， it has a certain phenomenon of misclassification， and the low spatial resolution affects the details of cloud boundary extraction results. The cloud detection performance of the Cloud-Score algorithm is obviously better than that of the QA60 method， the overall accuracy is 89.83%， and the commission error is only 2.17%， but there is still a phenomenon that some cirrus clouds are missed. Compared with the other three cloud detection methods， the cloud detection model proposed in this study has the highest overall accuracy， reaching 98.21%， and has extremely low omission error and commission error. The model can accurately identify the boundary of the cloud， and can meet the cloud detection preprocessing requirements of Sentinel-2 remote sensing products.