基于空间‑光谱级联差分网络的病理学高光谱图像融合方法

Pathological Hyperspectral Image Fusion Method Based on Cascaded Spatial‑Spectral Differential network

高分辨率高光谱病理学图像包含细粒度的二维空间信息和光谱维信息,对于精确诊断具有重要意义。现有方法存在空间光谱信息提取不充分和边缘信息利用不足的问题,为此,提出一种基于空间-光谱级联差分网络的病理学高光谱图像融合方法,通过级联空间-光谱模块重建高分辨率高光谱病理学图像,其中,每个空间-光谱模块都采用差分的方法分别设计了空间边缘损失和光谱边缘损失来分阶段约束和优化模型,从而实现病理学高光谱图像和多光谱图像的有机融合。该模型在病理图像数据集上进行实验,融合图像的4项评价指标RMSE,PSNR,ERGAS,SAM分别达到4.5937,32.3280,4.6683和3.6354,研究表明,基于空间-光谱级联差分网络的融合方法能够对细节较为丰富的病理学图像实现精细融合,为多光谱图像融合提供了参考依据。

High-resolution hyperspectral pathology images contain fine-grained two-dimensional spatial information and spectral dimensional information,which is of great significance for accurate diagnosis.To this end,we propose a pathological hyperspectral image fusion method based on the space-spectrum cascade differential network.In this method,high-resolution hyperspectral pathological images are reconstructed by cascading spatial-spectral modules.Each space-spectrum module utilizes differential methods to design spatial edge loss and spectral edge loss,which helps constrain and optimize the model in stages.This enables the organic fusion of pathological hyperspectral images and multispectral images.The model experimented on the pathological image dataset,and the four evaluation indexes RMSE,PSNR,ERGAS,and SAM of the fusion images reached 4.5937,32.3280,4.6683 and 3.6354,respectively.These results demonstrate that the fusion method based on the space-spectrum cascade differential network can achieve fine fusion of pathological images with rich details,providing a reference for multispectral image fusion.