【摘要】 云检测对于遥感图像的应用具有重要意义。目前已有的云检测方法关于遥感图像的偏振信息研究较少，性能和泛化能力有限。为有效利用遥感图像偏振信息，提出了一种基于深度学习的多模态融合遥感图像云检测方法并进行了初步实验评价。该网络是一种三参数输入流架构，具有编码器-解码器结构，利用通道空间注意模块对遥感图像中的反射率特征和偏振特征进行多模态融合。在解码器上采样阶段，利用迭代注意特征融合方法融合高、低级特征映射。评价实验数据集来源于多角度偏振成像仪(DPC)云产品和云掩码产品。评价实验结果表明，所提出的网络模型实现了良好的云检测性能，识别准确率达到93.91%。更多还原

【Abstract】 Cloud detection is of great significance for the application of remote sensing images. However,as for the existing cloud detection methods, there is limited research on the polarization information of remote sensing images, and their performance and generalization ability are also limited. To effectively utilize the polarization information of remote sensing images, a multimodal fusion remote sensing image cloud detection method based on depth learning is proposed and its preliminary experimental evaluation is conducted. In the method, the network is a three-parameter input stream architecture with an encoderdecoder structure, and the channel-spatial attention module is used to perfom multimodal fusion of reflectance and polarization features in remote sensing images. In the upsampling stage of the decoder, the iterative attention feature fusion method is used to fuse the high-and low-level feature maps. The evaluation experimental data set comes from Directional Polarization Camera(DPC) cloud products and cloud mask products. The evaluation results show that the proposed network model achieves good cloud detection performance, with a recognition accuracy of 93.91%.更多还原