基于决策树的戈壁信息提取研究

 戈壁是我国西北干旱地区的一种主要地表景观类型。由于戈壁所处地带自然条件恶劣，人烟稀少不便于实地调查研究的开展，所以目前关于戈壁的研究资料非常少。本文以戈壁典型分布区内蒙古自治区额济纳旗为例，以Landsat TM遥感影像以及30 m DEM作为基本数据源，首先借助DEM数据，在GIS中，提取出非戈壁分布区，然后，在分析不同典型地物光谱特征基础上，建立基于知识的决策树信息提取模型，利用该模型实现了侵蚀型戈壁与堆积型戈壁信息的自动提取，戈壁信息提取总体精度达到91.57%，研究为使用遥感影像快速获取戈壁信息提供了技术支持。

Gobi information extraction based on decision tree classification method

Gobi is one of the main landscape types of earth’s surface in the arid region of northwestern parts of China, with the total area of 458 000-757 000 km2, accounting for the 4.8%-7.9% of China’s total land area. The gobi holds abundant natural resources such as minerals, wind energy and solar power. Meanwhile, many modern cities and towns and some important traffic routes were also constructed on the gobi region. The gobi region plays an important role in the construction of western economy. Therefore, it is important to launch the gobi research under current social and economic conditions, and accurately revealing the distribution and area of gobi is the base and premise of launching the gobi research. At present, it is difficult to do fieldwork due to the execrable natural conditions and the sparse dweller in the gobi region, which leads to the scarcity of research documents on the situation, distribution, type classification, transformation and utilization of gobi. The studied region of this paper is a typical gobi distribution region, locating in Ejina County in Inner Mongolia, China, and its climatic characteristics include lack of rain, more evaporation, full sunshine, large temperature difference and frequent windy sand weather. Using Remote Sensing imageries Landsat TM5 and TM7 of plant growth season of 2005-2010, the DEM with 30 m spatial resolution, administrative map, present land use map, field investigation data and related documents as the basic data resource. Firstly, the non-gobi distribution regions were extracted in GIS software by analyzing DEM. Then, based on the analysis of spectral characteristics of difference typical ground objects, the information extraction model of Decision Tree based on knowledge was constructed to classify the remote sensing imageries, and eroded gobi and cumulated gobi were relatively accurately separated. The general accuracy of the extracted gobi information reached 91.57%. There were few materials in China on using remote sensing date to study the extraction of gobi information. This study relatively accurately realized the automatous extraction of eroded gobi and cumulated gobi based on the remote sending data, and these results would provide the technical support for obtaining the gobi information quickly from remote sensing images.