基于纹理和地形辅助的山区土地利用信息提取研究

发布时间：2018-05-01 05:32

本文选题：遥感 + 面向对象分类　；参考：《四川农业大学》2014年硕士论文

【摘要】：随着遥感传感器的飞速发展,遥感影像的空间分辨率得到大幅度的提升,地表成像细节变得更加清晰,但传统的遥感分类技术基于像元,不能充分利用高分辨影像丰富的图像特征,故其分类精度和准确性还不能满足土地调查的需要。山区土地利用类型多样,地形地貌复杂,土地利用分散,信息提取困难,如何采用合理有效的方法实现山区高分辨率遥感影像土地利用信息的准确、快速提取已成为土地利用调查中迫切需要解决的问题。本文选择位于川西南攀枝花市西区的典型农村山区为研究区,以全色波段分辨率为1m,多光谱波段分辨率为4m的IKONOS高分辨率遥感影像为数据源,基于ERDAS 9.2、ENVI4.8、eCognition8.0和ArcGIS 9.3等软件平台,采用面向对象分类方法,辅以纹理和地形因子的参与,最终实现研究区土地利用信息的准确、快速提取。主要研究结论如下：(1)通过最优指数法(Optimum Index Factor, OIF)对研究区IKONOS高分辨遥感影像最佳波段组合方式进行研究。结果表明,研究区遥感影像最优波段组合方式为2、3、4(GRN、RED、NIR)波段组合,其OIF值最大,达到102.88,该组合影像-地物色调反差较明显,较好的反映了研究区土地利用信息。(2)本文设定了三个分割层次(level1、Ievel2、level3)对研究区遥感影像进行多尺度分割,并通过影像特征分析法和均值方差法对每个层次的最优分割尺度进行了探讨。结果表明,level1层的最佳尺度为20,可较好的对山区居民点和裸地进行分割；level2层的最佳尺度为45,可较好的对山区坑塘、有作物耕地、无作物耕地、其他林地和园地进行分割；level3层的最佳尺度为80,可较好的对山区水库、河流、有林地、荒草地、道路进行分割。(3)依据香农信息熵(Shannon entropy)原理筛选出了Homo(同质性)Con(对比度)、Ent(熵)、Asm(二阶距)四个纹理指标辅助本文山区土地利用信息的提取研究,并对纹理辅助分割的效果进行了对比分析。结果表明,纹理参与分割过程很好的改善了图像分割的效果,纹理参与分割后的影像,同一类地物内部其分割后多边形破碎度明显降低,分割过程对大面积地物的边界信息予以充分考虑,地物的整体性得到体现。此外,多边形的数目明显减少,减少幅度为44.27%,无形中大大提高了影像解译的效率。(4)基于地形和纹理辅助的面向对象分类各精度评价指标较传统监督分类都有较大的提升,充分体现了该分类方法在山区土地利用信息提取中的优越性。其中,分类总体精度达到90.57%,较传统监督分类提高17.92%；Kappa系数值达到0.8892,较传统分类提高0.1879；各地类的生产者精度和用户精度较传统分类都有不同程度的提高；面向对象分类结果中各地类面积与实地调研地类面积更为接近,也较好的反映了该分类的准确性。
[Abstract]:With the rapid development of remote sensing sensors, the spatial resolution of remote sensing images has been greatly improved, and the surface imaging details become clearer, but the traditional remote sensing classification technology is based on pixels. The classification accuracy and accuracy of high resolution images can not meet the needs of land survey because they can not make full use of the rich image features of high resolution images. There are various types of land use, complex topography and geomorphology, scattered land use and difficult information extraction in mountainous areas. How to use reasonable and effective methods to realize the accuracy of land use information of high resolution remote sensing images in mountainous areas, Rapid extraction has become an urgent problem in land use survey. In this paper, a typical rural mountainous area located in the western part of Panzhihua City, Southwest Sichuan is selected as the research area. The IKONOS high-resolution remote sensing image with a panchromatic band resolution of 1m and a multi-spectral band resolution of 4m is used as the data source, and based on the software platforms of ERDAS 9.2 ENVI4.8e Cognition 8.0 and ArcGIS 9.3, etc. The method of object oriented classification, with the participation of texture and terrain factors, is used to realize the accurate and rapid extraction of land use information in the study area. The main conclusions are as follows: (1) the optimal band combination of IKONOS high-resolution remote sensing images in the study area is studied by the optimal exponential method of Optimum Index Factor (OIF-1). The results show that the optimal band combination mode of remote sensing image in the study area is the band combination of 2 ~ 3 ~ 3 ~ 4G ~ (?) r ~ (?), and its OIF value is the highest, reaching 102.88. The contrast between the image and the ground color is obvious. In this paper, we set up three levels of segmentation: level 1 and I level 2 level 3) to segment the remote sensing image of the study area by multi-scale. The optimal segmentation scale of each level is discussed by means of image feature analysis and mean variance method. The results show that the optimal scale of level 1 layer is 20, and the optimal scale of dividing level 2 layer of residential area and bare land in mountainous area is 45, and the optimal scale is 45 for pothole, cropland and no cropland in mountain area. The optimal scale of dividing three layers of forest land and garden land is 80, which can be better for mountain reservoirs, rivers, woodlands, and grasslands. According to Shannon's information entropy (Shannon entropyy) principle, four texture indexes of Homo (Ent) Ent (second order distance) were selected to help the extraction of land use information in mountain areas. The effect of texture aided segmentation is compared and analyzed. The results show that the texture involved in the segmentation process can improve the effect of image segmentation. After the texture is involved in the segmentation of the image, the degree of polygon fragmentation in the same kind of ground objects is obviously reduced after segmentation. In the process of segmentation, the boundary information of large-area objects is fully considered, and the integrity of objects is reflected. In addition, the number of polygons is obviously reduced by 44.27, and the efficiency of image interpretation is greatly improved.) the accuracy evaluation indexes of object oriented classification based on terrain and texture assistance are greatly improved than those of traditional supervised classification. The advantages of the classification method in the extraction of land use information in mountainous areas are fully demonstrated. Among them, the overall accuracy of classification reached 90.57, the value of Kappa coefficient increased 17.92% than that of traditional supervised classification, and the value of Kappa coefficient was 0.889 2, 0.1879 higher than that of traditional classification. In the result of object-oriented classification, the area of each area is closer to that of field investigation, which also reflects the accuracy of the classification.
【学位授予单位】：四川农业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F301.2;S127

【相似文献】