基于改进WGAN-GP的多波段图像同步超分与融合方法

Multi-Band Image Synchronous Super-Resolution and Fusion Method Based on Improved WGAN-GP

针对低分辨率源图像的融合结果质量低下不利于后续目标提取的问题,提出一种基于梯度惩罚Wasserstein生成对抗网络(WGAN-GP)的多波段图像同步超分与融合方法。首先,基于双三次插值法将多波段低分辨率源图像分别放大至目标尺寸;其次,将该放大结果输入特征提取(编码)网络分别提取特征并在高层特征空间进行组合;然后,通过解码网络重构出初步融合图像;最后,经过生成器和判别器的动态博弈得到高分辨率的融合图像。实验结果表明,所提方法不仅可以同步实现多波段图像的超分和融合,而且融合图像的信息量、清晰度和视觉质量明显高于其他代表性方法。

Aiming at the problem that the fused results of low resolution source images are not good for the subsequent target extraction, a multi-band image synchronous super-resolution and fusion method based on Wasserstein generative adversarial network with gradient penalty(WGAN-GP) is proposed. Firstly, the multi-band low-resolution source images are enlarged to the target size respectively based on the bicubic interpolation method. Secondly, the enlarged results are input to a feature extraction(encoding) network to extract features respectively and combine them in a high-level feature space. Then, the initial fused images are reconstructed by decoding network. Finally, a high-resolution fused image is obtained through a dynamic game between the generator and the discriminator. The experimental results show that the proposed method can not only achieve multi-band images super-resolution and fusion simultaneously, but also the information amount, clarity, and visual quality of the fused images are significantly higher than other representative methods.