基于光学和雷达图像的土地覆被分类

Land-cover Classification Based on HJ1B and ALOS Data

为寻求一种有效的提高多源遥感数据土地覆被分类制图精度的方法,探讨了融合HJ1B和ALOS/PALSAR图像进行遥感图像分类制图的方法。在对光学图像HJ1B和雷达遥感数据ALOS/PALSAR进行离散小波融合的基础上,应用分类决策树CART(Classification and Regression Tree)算法对融合的图像进行了土地覆被分类制图,并将其分类结果与支持向量机SVM(Support Vector Machine)分类结果进行对比。研究结果表明:将光学和雷达图像数据进行离散小波融合,采用分类决策树CART和支持向量机SVM进行图像分类,CART的分类精度要优于SVM的结果。可见,在光学图像HJ1B和雷达数据ALOS/PALSAR融合的基础上,应用CART能有效进行地物识别,提高图像的分类精度。

In order to increase the accuracy of the land use and land cover （LULC）classification via multi-source remote sensing data,we explored an effective algorithm by fusion of HJ1B images from optical sensors and ALOS /PALSAR data from radar remote sensing.In the process of fusion,the discrete wavelet transform （DWT）was uti-lized.The land-cover classification mapping was performed by using the classification and regression tree （CART） approach.The classification result by CRT approach was compared with that by support vector machine （SVM）ap-proach.The results show that：1）through fusing HJ1B optical images with ALOS /PALSAR radar data,we obtain an overall Kappa coefficient （0.826 9）and total accuracy（85.60 %）by CRT approach,while by SVM approach the value is 0.816 7 and 84.82 %,respectively;2）in terms of classification accuracy,CRT approach is superior to SVM approach;3）by means of fusing optical images with radar data ,we can effectively carry out object recogni-tion and improve classification accuracy through applying CART approach.