基于多级特征补偿的遥感图像时空融合方法

在许多地球科学应用中要用到大量的高时空分辨力的地球观测数据。时空图像融合方法为产生高时空分辨力的数据提供了一种可行且经济的解决方案。然而，现有的一些基于学习的方法对于图像深层特征提取能力较弱，对于高分辨力图像细节特征利用度不够。针对这些问题，提出一种基于多级特征补偿的遥感图像时空融合方法。该方法使用2个分支进行多层级的特征补偿，并提出了融合通道注意力机制的残差模块作为网络的基本组成单元，可以将高分辨力输入图像的深层特征更为详尽地提取利用。提出一种基于拉普拉斯算子的边缘损失，在节省预训练计算开销的同时取得了很好的融合效果。使用从山东和广东2个地区采集的Landsat和中分辨力成像光谱仪(MODIS)卫星图像对所提出的方法进行实验评估。实验结果表明，提出的方法在视觉外观和客观指标方面都具有更高质量。

Spatiotemporal fusion of remote sensing images based on multi-level feature compensation

A large amount of earth observation data with the high spatial and temporal resolution is employed in many earth science applications. The spatiotemporal image fusion method provides a feasible and economical solution for generating high spatiotemporal resolution data. However, some of the existing learning-based methods are poor in extracting deep image features and utilizing the detail features of high-resolution image. A spatiotemporal fusion method is proposed for remote sensing images based on multi-level feature compensation. It uses two branches to perform multi-level feature compensation and proposes a residual module fused with a channel attention mechanism as the basic unit of the network, which can extract and utilize the deep features of high-resolution input images in more detail. An edge loss is proposed based on the Laplacian operator, which saves the computational cost of pre-training and achieves a good fusion effect. The proposed method is experimentally evaluated by using Landsat and Moderate-resolution Imaging Spectroradiometer(MODIS) satellite images collected from two regions in Shandong and Guangdong. Experimental results show that the proposed method bears higher quality in both visual appearance and objective metrics.