With the development of remote sensing technology, feature extraction methods are gradually diversified, mainly through optical and radar remote sensing. Optical remote sensing images contain a wealth of spectral information, whereas radar remote sensing images are all-day and all-weather. But they have some drawbacks. As a result, different features of Sentinel-1A and Landsat-8 images are combined in this paper to exploit their advantages for feature extraction experiments fully. Data preprocessing for Sentinel-1A and Landsat-8 is performed in this paper, followed by polarization feature extraction using H-α-A decomposition for Sentinel-1A and texture feature extraction using GLCM for Sentinel-1A and Landsat-8, followed by feature combination and classification using SVM classifier. Finally, the accuracy of classification results is evaluated. The results of this paper are as follows: the worst accuracy result is based solely on Landsat-8 spectral features combination, with an overall accuracy of only 80.61% and a Kappa coefficient of 0.6702; the accuracy of features combinations based on Landsat-8 spectral plus texture and Sentinel-1A polarization plus texture is improved. The best accuracy is 90.78%, and the Kappa coefficient is 0.8473. The experimental results show that multi-source remote sensing-based feature extraction with multiple feature combinations is more advantageous.
|