基于改进型U-Net的遥感云图分割方法

云检测是遥感图像预处理的重要步骤，云检测精度直接影响后续遥感图像应用的准确性，针对现有云与云阴影分割任务中，泛化能力差，误检漏检现象严重的问题，本文提出了一种改进型U-Net网络模型，该模型以U-Net为主干网络，加入高效通道注意力机制，修改激活函数。将遥感图像作为输入，放入基于高效通道注意力的U型云图分割模型中进行训练，在获得最优权重后，输出包含云区域、云阴影区域和背景区域的遥感图像分割结果。实验结果表明，相比于现有分割模型，本模型在云和云阴影的分割任务中参数量最低，准确率最高，泛化效果最好。

Method of remote sensing cloud image segmentation algorithm based on improved U-Net

Cloud detection is an important step in the preprocessing of remote sensing images. The accuracy of cloud detection directly affects the accuracy of subsequent remote sensing image applications. Aiming at the problems of poor generalization ability and serious misdetection and missed detection in the existing cloud and cloud shadow segmentation tasks, this paper designs an improved convolutional neural network model, which uses U-Net as the backbone network and adds efficient channel attention mechanism and modifies activation function. The remote sensing image is used as input and put into the U-shaped cloud image segmentation model based on efficient channel attention for training. After obtaining the optimal weight, the remote sensing image segmentation result including cloud area, cloud shadow area and background area is output. The experimental results show that, compared with the existing segmentation models, this model has lowest parameters, highest accuracy, and best generalization effects in the segmentation task of clouds and cloud shadows.