多层级图像融合算法用于水体探测的比较

Comparative Study on Water Detection Algorithms Based on High-resolution Optical and SAR Imagery Fusion

针对基于多源遥感图像的水体探测算法,缺少不同层级融合算法对水体探测精度的比较分析的问题,开展了比较不同级别的融合算法对水域面积提取精度的影响分析。分别在像素级与特征级对高分辨率光学影像与合成孔径雷达图像进行融合,并通过Kompsat-3及Kompsat-5图像开展水体探测实验。像素级融合算法基于IHS色度空间变换;特征级融合算法基于最大后验概率-马尔可夫随机场融合框架。基于ROC曲线对实验结果进行了定量比较分析。实验结果表明基于IHS色度空间变换算法的像素级图像融合方法水体探测精度与基于归一化差分水体指数的水体探测精度接近,而基于马尔可夫随机场算法的特征级图像水体探测方法综合利用了光学影像与合成孔径雷达图像上的水体特征,拥有区分水体与阴影的能力,水体探测精度较高。

The comparative analysis of water detection accuracy by fusion algorithm of multi source remote sensing images between different fusion levels is still scare.Therefore,the effect of fusion algorithms in different fusion levels on the water detection accuracy is conducted.Fusion algorithms in both pixel and feature levels are used to fuse high resolution multi spectral optical image and SAR data,and were tested on Kompsat3and Kompsat5data.The pixel level fusion algorithm is based on the IHS transformation;while the feature level fusion algorithm is based on the maximum posterior Markov Random Field fusion framework.The ROC curves are used to compare the fusion algorithms quantitatively.The result shows that the IHS transformation based pixel level fusion algorithm presents similar accuracy as the NDWI image in water detection;while the MRF based feature level fusion algorithm combines the different water characters in optical and SAR images,exhibits the ability to distinguish water from shadow regions,and achieves higher accuracy of water detection.