Abstract:
Using Landsat 8 OLI image data of the same region and different phases, 24 characteristic variables, such as spectral, texture and geography, are used to study the image classification with random forest classification(RF) and support vector machine classification (SVM) in Shizuishan area in Ningxia. The results show that there are some important parameters such as DEM data, normalized difference vegetation index(NDVI), shortwave infrared band, normalized difference humidity index (NDMI) and first principal component mean (M),affecting the accuracy of image classification. The classification accuracy of RF is slightly higher than that of SVM, the overall classification accuracy is 95.492%, Kappa coefficient is 0.947. The accuracy of saline-alkali soil category is 98.510%, and the computational efficiency is 16.5 times of SVM. RF is more suitable for detecting the saline land change at target level. According to the RF classification results of the two time-series images, the area of saline-alkali land decreased by 133.56 km2 from 2014 to 2017, with a total reduction of 56.368%. The improvement of ecological environment and the improvement of saline-alkali land are better.
