我国从1984年开始，先后进行过两次土地大调查。两次调查历时十多年，耗资、耗时、耗人力，主要原因在于面对遥感图像多数据源、多时相及大数据量造成的复杂问题，难以解决计算机解译分类，而采用野外实地核查为主的土地调查分类方法，其中工作的艰苦还不算最大问题，因时间跨度很大，数据的时效性都难以保证，因此，遥感图像解译的传统方法已面临巨大挑战。遥感科学与技术的快速发展，应用遥感图像进行土地利用现状调查无疑是最为科学、有效的方法。在目前的遥感分类应用中，用得较多的的是传统的模式识别分类方法，其分类结果由于遥感影像本身的空间分辨率以及“同物异谱”、“异物同谱”现象的存在，而往往出现较多的错分、漏分情况，导致分类精度不高。 本选题的研究目标是完善遥感图像的计算机解译理论方法，应用计算机自动、高效地完成土地调查中的地类划分问题，特别是要解决难度最大的林地、草地、园地和耕地等绿色植被的地类划分问题。具体的工作是： （1）对地物自动提取的可行性进行分析，将现有数据做感兴趣区域进行分析，为后面的图像分类做数据准备。 （2）为了进一步提高分类结果，选用频率域增强中同态滤波、空间域增强中直方图均衡、辐射增强中去霾方法，对遥感影像进行了预处理，在一定程度上提高了影像中地物的识别度。 （3）论文通过分析目前比较流行的地物信息自动提取方法，在全面分析研究区影像的光谱值、纹理结构等特征的基础上，构建了研究区地物的新的分类规则。 （4）通过得到的新规则树，对影像进行决策树分类。发现结合像元与纹理信息的分类方法比传统的基于像元的分类方法准确性高，特别是对有“同物异谱”、“同谱异物”现象的影像进行地物信息的自动提取时具有绝对优势。

Our country has had two national land investigations since 1984. Both surveys for more than ten years,but based on the principle land investigation taxonomic approach of open country investigate, the difficult work does not the most major problem, as the time span is very big, the data effectiveness with difficulty guaranteed, therefore, the conventional routes of remote sensing image interpretation has faced with huge challenge. Along with the development of computer imagery processing technology and remote sensing technique, the research and the application of the terrain feature of automatic extraction of remote sensing image obtains people's favor more and more. Take the remote sensing image pretreatment as the foundation, realizes automatic extraction of the remote sensing image terrain feature, gains the different topic information of the surface fastly, and uses these special information renewal geography database fastly, it is not only one of the basic researches of remote sensing image automatic understanding, but also is the benefit efficient path which can increase the precision and the work efficiency.While the “same thing but different spectrum” and “same spectrum but foreign matter” phenomenon still exist. (1) The feasibility of automatic extraction of surface features analysis of existing data to do analysis of regions of interest for subsequent image classification for data preparation. (2) In order to further improve the classification results, the choice of frequency domain enhancement homomorphic filtering, histogram equalization, spatial enhancement, radiation to the haze in the method, the preprocessing of remote sensing images, to a certain extent, to improve image Identification of material degree. (3) The paper analyzes the current surface features popular automatic extraction of information, a comprehensive analysis of the study area in the spectrum of the image, texture features based on the structure, constructed in the study area features on the new classification rules. (4) Get the new rule tree, decision tree classification of images. Found that combining pixel classification and texture information than the traditional pixel-based classification method with high accuracy, especially for the "same object different spectrum," "with the spectrum of foreign body" in the images when the surface features automatic extraction of information with absolute advantage.