Land Cover Classification of Huixian Wetland Based on SAR and Optical Image Fusion

Abstract

In this paper, GF-1 WVF image and Sentinel-1 SAR image covering Huixian wetland area are used as data sources. The Gram Schmidt (GS) algorithm is first used to fuse GF-1 images and SAR images with different polarization modes, and then the Random Forest (RF) algorithm is used for supervised classification. Finally, the accuracy of classification results and the ability to extract information are compared. The experimental results show that the fusion image has obvious texture features and prominent karst landform features, compared with the GF-1 WVF image. Compared with the Sentinel-1 SAR image, the fusion image has obvious spectral features. Spectral differences between typical features are large; The overall classification accuracy of GF-1 images, GF-1 and Sentinel-1 VV polarization fusion images, and GF-1 and Sentinel-1 VH polarization fusion images have reached over 80%. The classification accuracy of GF-1 and Sentinel-1 VV polarization fusion images reaches 85.15%, which is better than GF-1 and Sentinel-1 VH polarization fusion images. The classification accuracy of water bodies in the VV polarization fusion image is better than that of GF-1. Bare ground has the highest classification accuracy among all fused images.