Based On China-Made GF-1 For Land Use/Cover Classification Of High Cold Mountain Areas

The changes of Land use/cover is one of the core contents in current global change research, and land use/cover classification is one of its basic work. Recently, the remote sensing technology develop rapidly, such as the high-resolution GF-1 satellite appeared, more and more high-resolution satellites were launched and used in China. Using remote sensing technology to obtain the land use/cover classification has become an effective tool and techniques. And there is little reference for land use/cover classification in alpine mountains using the high-resolution GF-1 satellite. This paper selected the typical alpine mountain areas which is in Bayi town of Linzhi areas in Tibet using the high-resolution GF-1 satellite to make a preliminary inquiry for land use/cover classification. The main work and conclusions are as follows:(1) This paper using the pixel-based maximum likelihood supervised classification, divisional ISODATA unsupervised classification and object-oriented classification to get the land use/cover classification of Bayi town of Linzhi areas in Tibet with the WFV2 data of GF-1 data for basic data and combined with the 2m panchromatic data of GF-1. These results could provide some reference for land use/cover classification in alpine mountain areas using the high-resolution GF-1.(2) With making a reference for Chinese land use/cover classification system and American land use/land cover classification system and combining with the actual survey for the real surface features and the complex alpine mountain terrain for this areas, the land use/land cover for this study area make into water body, vegetation, town and road, farmland, bare land and unutilized land in six broad categories.(3)To solve the wrong problems for water body and hillshade due to terrain factors in this area using the ISODATA unsupervised classification, we make a boundary in the 3050m contour after many trials combined with the complex terrain in this study areas and this study areas was divided into less than 3050m above sea level and an altitude of more than 3050m regional area to divisional ISODATA classification which solved the problem of the misclassified for the water bodies and the hillshade.(4)In object-oriented classification, we make a study for the best segmentation scale in our study area. We used the ESP algorithm (Estimation of scale parameter) to calculate the initial three best segmentation scale for this study area are 50,70 and 90 pixels. And then we made a comparison with the above three best segmentation scale using a KNN classification for the land use/cover classification’s accuracy of this study area.(5) We made an accuracy assessment for this study area land use/cover classification using maximum likelihood classification, divisional ISODATA classification and object-oriented classification. And we got the result that the accuracy of object-oriented classification using the best segmentation scale 50,70 and 90 pixels are all better than the traditional maximum likelihood supervised classification and ISODATA unsupervised classification. And when the best segmentation scale is 50 pixels in objected-oriented classification, the overall classification accuracy is 84.87% which is increased by 21.69% for the traditional maximum likelihood classification overall classification(63.18%) and 34.72% for the ISODATA unsupervised classification overall classification(50.15%).