Mountainous Land Use Information Extraction Based On Texture And Terrain-Assisted

With the rapid development of remote sensing sensors, spatial resolution remote sensing image has been greatly enhanced, surface Imaging details become more clearly, but the traditional techniques of remote sensing classification based on primitive, can not take full advantage of high-resolution remote image feature-rich, so the classification precision and accuracy can not meet the needs of land survey. Mountain land use type variety, complex topography, land use decentralized, land use information extraction is difficult, how to adopt reasonable and effective method to realize land use information extraction accurately and quickly from the mountains high resolution remote sensing image has become an urgent need to solve problems in land use investigation. In this paper, a typical rural mountain located in the western of southwestern Sichuan Panzhihua was chose for the study area, with a resolution of 1 m panchromatic band, multispectral band resolution for 4 m IKONOS high-resolution remote sensing image as data sources, based on software platform of ERDAS 9.2, ENVI 4.8,8.0 eCognition and ArcGIS 9.3, using object-oriented classification method, supplemented by texture and terrain factors, finally, accurately and quickly realized the land use information extraction. The main research conclusions were as follows:(1) Through the Optimum Index Factor (OIF) method to study the optimal band combination of IKONOS high-resolution remote sensing image in the studied area. Results showed that the optimal band combination way as the remote sensing image in the study area was 2,3,4 (GRN, RED, and NIR) band combination, its biggest OIF values, at 102.88, the composite image feature tonal contrast was more apparent, better reflect the information of land use in the study area.(2) This study set the three split levels (level1, level2, level3) for remote sensing image multi-scale segmentation in study area, and through the image feature analysis method and mean-variance method for the optimal segmentation scale of each level were discussed in this paper. Results showed that the optimal segmentation scale of level 1 was 20, residential areas and bare land can be better extracted in mountainous area; The optimal segmentation scale of Level 2 was 45, ponds,crop cultivated land,no-crop cultivated land,other woodland and garden can be better extracted in mountainous area; The optimal segmentation scale of level 3 was 80, reservoir,river, road, forestland and grassland can be better extracted in mountainous area.(3) Based on Shannon entropy principle to filter out four texture index information which were the Homo (homogeneity) Con (contrast), Ent (entropy), Asm (second moment) to assist extraction of mountainous land use, and texture assisted segmentation results were compared and analyzed in this article. The results showed that the texture to participate in the segmentation process was very good to improve the effect of the image segmentation, texture in the image after segmentation, the same features internal polygon fragmentation decreased obviously after the segmentation, the segmentation process of large area of the boundary of the object information should be fully considered and the integrity of the object. In addition, the number of polygons decreased significantly, decrease rate of 44.27%, virtually greatly improve the efficiency of image interpretation.(4) Each evaluation index which used object-oriented classification based on the terrain and texture auxiliary had larger improvement than traditional supervised classification accuracy, fully embodied the superiority of classification method to extract information of land use in the mountains. Among them, the overall classification accuracy reached 90.57%, compared with traditional supervised classification increased by 17.92%; Kappa value reached 0.8892, compared with traditional supervised classification increased by 0.1879; Producers of accuracy and precision than traditional consumers classification had different degree of increase; The area of around the class results which used object-oriented classification were closer to field survey class area, but also better reflected the classification accuracy.