Study On Methods Of Extraction Of Utilization Information Of Land Based On SPOT5

This study took Changjiaba town of Foping county in Shaanxi province as the test area. The SVM classification model combined with texture analysis based on texture extraction from SPOT5 remote sensing images with the utilization of RS and GIS analysis software ENVI4.7 and ArcGIS9.2. In this paper, the authors used distance-based approach J-M to calculate the separability, and then estimated the optimal separated scales, consequently, the texture information obtained under different scales participated in the multi-scale image classification. With the support of ENVI ZOOM, the object-oriented classification and multi-scale segmentation was used to process images. After that, we combined spectrum information, NDVI and texture information to extract land utilization information from the test area. Finally, we compared the results of different classifications, for the purpose of providing an evidence for studying land utilization and cover dynamic variation. The conclusion of this research is listed as follows:(1) Using the distance-based approach J-M to calculate the optimal separated scales of different land types. After analyzing we concluded that the optimal separated scale of unused land, water, arable land, road, forest land and building plots is 3×3, 5×5, 7×7, 11×11, 13×13 and 15×15.(2) This study integrated the multi-window texture features and spectrum characteristics to classify the remote sensing images. The results indicate that such method is obviously more accurate than that solely with spectral information and also can extract land utilization information more precisely. Consequently, this classification has higher accuracy and Kappa coefficient.(3) The classification with participation of multi-texture, to some extent, can improve the accuracy and results. Specifically, the classification accuracy with three-window texture (3×3, 13×13 and 15×15) is distinctly higher than that with single one, with the overall accuracy and Kappa coefficient is 78.50% and 0.7380, respectively. The result suggested that such classification could not only distinguish land objects on a same scale more easily, but also solve the problem of same spectral from different materials and same material with different spectral.(4) The object-oriented classification could effectively improve the accuracy, explicitly, the overall accuracy and Kappa coefficient is 84.50% and 0.8113 respectively, which severally increased 6% and 0.0733 compared with SVM classification.(5) With the participation of texture characteristics, multi-scale segmentation of object-oriented classification can efficiently reduce spectral heterogeneity of land objects, which expresses more reasonable while identifying land objects.