云检测是遥感影像处理的重要任务之一。目前遥感影像云检测中多使用到卫星的多光谱、多通道信息, 而关于多角度信息对云检 测影响的研究较少。为了探索遥感影像多角度信息作为云特征对训练云分类网络精度的影响, 提出一种基于深度学习的遥感多角度 云检测方法, 以 SegNet 为基础网络结构, 提取含有多角度信息的遥感影像的特征表示, 训练含有多角度信息的遥感影像云检测模型。 测试结果表明, 所提方法全局精度为 91.39%, 平均重叠率为 83.99%。分析表明单角度云检测具有一定的局限性, 而利用多角度信息 作为云特征训练云分类网络可以提升云检测精度。此外, 还探索了 POLDER 仪器中不同角度对于云检测结果的影响情况。

Cloud detection is one of the important tasks for remote sensing image processing. At present, the multi-spectral and multi-channel information is often used in cloud detection of remote sensing image, but the research on the influence of multi-angle information on cloud detection is still insufficient. To explore the influence of multi-angle information as cloud feature on the accuracy of cloud classification, a cloud detection method with multi-angles remote sensing based on deep learning is proposed. The proposed method takes SegNet as backbone network, and trains a multi-angle information based cloud detection model by extracting the remote sensing image feature with multi-angle information. Extensive experimental results demonstrate that the Global Accuracy and the mean intersection over union (MeanIoU) of the proposed method are 91.39% and 83.99% respectively. And the method proves the limitations of single angle cloud detection and the effectiveness of multi-angle information on the improvement of the cloud detection accuracy. In addtion, the influence of different angles on the cloud detection in POLDER is also explored.