基于光学和SAR遥感图像融合的洪灾区域检测方法

基于光学和合成孔径雷达(SAR)图像融合的洪灾区域检测方法可以全天候、高时效地检测洪灾区域。由于SAR图像中存在大量随机分布的相干斑噪声,传统洪灾区域检测方法的检测结果存在较高的虚警率。该文在模糊C均值聚类方法(FCM)的基础上提出了分级聚类算法(H-FCM),该方法将洪灾后的SAR图像与洪灾前的光学图像融合。基于融合图像,该方法利用提出的分级聚类模型获得洪灾区域的初步检测结果。此外,该算法在利用所提出的区域生长算法获得洪灾前河流位置后,将其作为初步检测结果的空间约束,进一步筛除疑似洪灾区域,并显著地提升了检测性能。该文的实验数据包括1999年英国格洛斯特洪灾前后的遥感图像和2019年中国南昌洪灾前后的遥感图像。通过对比实验,H-FCM算法的有效性得到验证。

Flooded Area Detection Method Based on Fusion of Optical and SAR Remote Sensing Images

The flooded area detection method based on the fusion of optical and Synthetic Aperture Radar (SAR) images is applicable for all weather conditions and times. However, due to the large number of randomly distributed intensive speckle noise in SAR images, the conventional methods of detection often trigger high false alarm rates at flood-stricken zones. Inspired by the Fuzzy C-Means (FCM) clustering method, a hierarchical clustering algorithm (Hierarchical Fuzzy C-Means, H-FCM) is proposed in this paper. This method fuses the SAR image captured after the flood with the optical image captured before the flood. Based on the fused image, this method uses the proposed hierarchical clustering model to obtain the preliminary detection results of the flooded area. Additionally, the algorithm uses the proposed region-growing algorithm to obtain the river location before the flood and uses it as a spatial constraint for the preliminary detection results to further screen out suspected flooded areas and significantly improve detection performance. The experimental data used in this paper include the remote sensing images captured before and after the Gloucester floods in the United Kingdom in 1999, as well as the remote sensing images captured before and after the Nanchang floods in China in 2019. The effectiveness and validity of the H-FCM algorithm are also supported by comparison experiments.