基于跨层复制连接卷积神经网络的遥感图像融合

首先, 基于卷积神经网络提出一种采用跨层复制连接操作融合不同尺度特征图的遥感图像融合模型, 解决了传统遥感图像融合方法对不同类型遥感图像需人为选择不同的分解融合规则, 导致融合图像质量受选择规则影响较大的问题. 其次, 使用Deimos卫星和QuickBird 卫星数据验证该方法的有效性, 并用主观和客观相结合的方法评价融合图像质量. 实验结果表明, 该遥感图像融合模型与传统方法相比, 能有效将全色图像的空间信息与多光谱图像的光谱信息融合, 并抑制光谱扭曲.

Remote Sensing Image Fusion Based on Cross LayerCopy Connection Convolutional Neural Network

Firstly, based on the convolutional neural network, we proposed a remote sensing image fusion model that used cross layer copy connection operations to fuse feature maps of different scales, and solved the problem that traditional remote sensing image fusion methods needed to manually select different decomposition and fusion rules for different types of remote sensing images, w hich led to the problem that the fusion image quality was greatly affected by selected rules. Secondly, the effectiveness of the method was verified by the Deimos satellite and QuickBird satellite data, and the fusion image quality was evaluated by the subjective and objective methods. Experimental results show that the proposed model can effectively combine the spatial information of panchromatic image with the spectral information of multispectral image, and control the spectral distortion.