摘要
以加拿大Radarsat SAR与美国Landsat TM影像为信息源,分别将SAR与TM影像的DN值转换为表征地物特征的后向散射系数和反射率,利用改进的SVR法进行融合,同时与HIS,Brovey以及小波变换的融合效果作定量比较,并利用优化的BP神经网络模型,以相同的训练区分别对融合前后的影像进行监督分类。结果表明:改进的SVR法融合影像的光谱信息保持性、信息量以及分类精度都优于常用的融合方法,且分类精度比TM影像有较大提高。
The Radar SAR and Landsat TM images are selected as the experimental datum, after converting the DN value to backscatter coefficient and reflectance respectively, the improved SVR method is used to fuse the data. To compare with fusion result, the HIS, brovey and wavelet method are used in this article. Finally, the optimized back propagation model is used to classify the TM and the fusion images with the same training areas. The experiment results are analyzed in spectrum maintenance or in quantity and the accuracy of the classification, it shows that the improved SVR fusion method is superior to other commonly used fusion methods.
出处
《测绘学报》
EI
CSCD
北大核心
2006年第3期229-233,239,共6页
Acta Geodaetica et Cartographica Sinica
基金
SustainableAgroecosystemManagementandDevelopmentofRural-UrbanInteractioninregionsandcitiesofChina(SUSDEV-CHINA)(ICA4-CT-2002-10004)
