面向特征识别的SAR与可见光图像融合算法研究

Fusion algorithm of SAR and visible images for feature recognition

合成孔径雷达(synthetic aperture radar,SAR)图像与可见光图像由于成像机理的不同,图像差异较大,传统融合算法不利于地物目标特征的解译与识别,难以满足应用的需求。基于此,文章结合SAR与可见光图像的成像特征,提出了一种基于相位一致性特征检测的自适应HIS(hue-intensity-saturation)融合算法,该算法首先利用相位信息区分目标轮廓和纹理边缘,提取SAR图像的空间特征信息,能有效地避免过量纹理边缘的提取。自适应HIS算法能调节2个图像之间的灰度相关性,实现加入信息的平滑过渡,减少光谱扭曲。实验以哨兵1号C波段SAR图像与Landsat8可见光图像进行验证,并与传统的HIS、主成分分析(principal component analysis,PCA)、小波以及目前较为主流的NSCT(nonsubsampled contourlet)算法进行比较,结果表明:该算法能有效地加入可见光图像不易察觉的特征信息,实现了融合图像在光谱保持和空间结构与特征保持上较好的权衡,提高了图像的地物检测与目标识别的能力。

The large difference between synthetic aperture radar(SAR)and visible images due to the different imaging mechanism makes it more difficult for the general fusion methods to interpret and recognize the features of ground targets.In this paper,in view of the imaging characteristics of SAR and visible images,the fusion method based on the adaptive hue-intensity-saturation(AHIS)and phase congruency feature detection is proposed.Firstly,the phase congruency is used to distinguish target contour and texture edge and extract the spatial features of the SAR image,which inhibits the texture edge effectively.Then the AHIS is used to adjust the gray correlation between two images to get a smooth injection of SAR image features.A fusion of Landsat8 visible image with Sentinel-1A SAR image is made to validate the performance of the proposed method.Experimental result shows that compared with hue-intensity-saturation(HIS),principal component analysis(PCA),wavelet and nonsubsampled contourlet(NSCT),the proposed method can effectively inject the missing features into the visible image and better maintain the spectral and spatial structure information,thus improving the capability of image target detection and recognition.