基于光谱梯度角的改进SREM融合方法

An Improved SREM Fusion Method Based on Spectral Gradient Angle

针对多光谱图像光谱分辨率增强方法(Spectral Resolution Enhancement Method,SREM)在重采样以及空间配准过程中出现的像元位置和信息量偏差,利用同名点匹配的思想,计算高光谱影像上像元光谱,与多光谱影像上对应位置及其邻域内像元光谱的光谱梯度角,获得最具相似特性的一对像元光谱组成光谱数据集,作为模型的输入数据,提出改进后基于光谱梯度角的ISREM-SGA光谱融合方法(Improved Spectral Resolution Enhancement Method-Spectral Gradient Angle,ISREM-SGA)。以开源图像Chikusei、Pavia U为基础实验数据,通过空间与光谱降维模拟高光谱和多光谱图像进行融合研究,并对融合结果进行质量评价。实验结果表明,改进后的ISREM-SGA提高了融合图像质量,具有较强的稳健性。Chikusei数据的相关系数R与通用质量评价指标UIQI值分别提高了0.03,相对无量纲全局误差ERGAS值降低了3.33;Pavia U数据R值提高了0.24,UIQI值提高了0.30,相对无量纲全局误差ERGAS值降低了35.82,说明了本研究改进后ISREM-SGA高光谱-多光谱遥感图像融合方法的有效性。

In order to solve the problem that image resampling and registration may cause the biased pixel due to the diversity and complexity of images,which affects the quality of SREM result,an improved spectral resolution enhancement method based on spectral gradient angle(ISREM-SGA)is proposed.“Pure pixels”are choosed based on spectral gradient angle between hyperspectral spectrum and multispectral spectrum,as parameters of model.Hyperspectral-multispectral image transform matrix are calculated,which generate the final fusion image.With Chikusei and Pavia U images as basic data,hyperspectral and multispectral images are simulated through cropping and resampling.Qualitative and quantitative methods are taken to evaluate the quality of fusion images.The results show that ISREM-SGA is more effective and steady than original method.In the ISRME-SGA method,the correlation coefficient(R)and the general quality evaluation index(UIQI)of Chikusei fusion image are increased by 0.03 and errors relative global adimensionnelle synthese(ERGAS)is decreased by 3.33;in the Pavia U fusion image,R is increased by 0.24,the UIQI is increased by 0.30,and the ERGAS is decreased by 35.82.This study illustrates the effectiveness of Improved Spectral Resolution Enhancement Method.