Remote Sense Retrievalof 30m Resolution Of Land Use Based On The Combination Of Unsupervised Classification And Decision Tree

Land is one kind of precious and limited resources to which human being’s survival and development is attached. It is also an important means of production and livelihood for human beings because all social and economic goals are realized through long-term operation on land. In 21 st century, the majority of land over the world has been exploited. Under such circumstance, of particular important is how to make full use of land to create more material wealth and better serve the economy.With everlasting development of human beings’ technology, cognition level towards properties contained in land deepens as well. Based on the knowledge adjusting land use paradigm can gain better social and economic profits. Therefore, land use is a dynamic process with the manifestation of material, energy and information’s communication and exchange between human being and land. Present analysis of land utilization is the basis of reasonable utilization of land resource. It is of significant meaning of improving the efficiency and quality on information extraction of current land use.Traditional method of obtaining land use type is field investigation. It is precise but takes too much human and material resource and cannot meet the rapid changes of time. Starting from 1970 s of remote sensingsatellite lift-off, the land use investigation research with remote sensing image began. With the features of mass information and comprehensive data, remote sensing image plays a important role in classifying land use. Due to difference of world climate, topology, hydrology, vegetation, social and economic situation, there isn’t a unified classification system and automatic interpretation scheme. The research spot falls intoclassification system and automatic interpretation scheme based on different field situations.The article selects Qinhuangdao region typical for Jingjinji area in northern China and refers to USGS、NLCD、IGBP、FAO, Chinese Academy of Science’s land use classification system and OIL sense feature fitting for land use classification system in reality. The land use type is divided into cultivated land, forestry land, rosebush, grassland, bare land, buildings and waters.Main methods of remote sensing classification are artificial visual interpretation,supervised classification and unsupervised classification, object-oriented, decision tree and so on. The paper integrates and links these traditional methods and automatically interprets combined with unsupervisedclassification and decision tree. The research has prepared data such as Landsat 8 OIL remote sensing image, NDVI Normalized Difference Vegetation Index, EVI Enhanced Vegetation Index, DEM Digital Elevation Model, slope chart, aspect chart, topographic feature chart and so on.First, unsupervised classification is used to have a general classification of remote sensing image and analyze integration of spectrum values of various types and forms land utilization type chart based on unsupervised classification. On such basis, 100 samples areas from the features of NDVI、EVI、altitude、slope、aspect、topographic feature are selected and credibility of separable sample area is verified. Then it is needed to calculate classification standard of decision tree based on C4.5 model of sample area. The last step is to execute decision tree and get classification result and assess the accuracy through matrix confusion. Based on the combined unsupervised classification and decision tree, the overall accuracy is about 86% and Kappa coefficient is about 0.80. That can meet the requirements of the National Natural Science Foundation of Jingjinji and atmospheric particulate matter concentration based on land use(41471091). It proves that the advancement of the methodology the paper uses and explains that the result is in accordance with true status of Qinhuangdao land surface type.The innovation points are as follows:(1)Combined USGS、NLCD、IGBP、FAO, Chinese Academy of Science’s land use classification system, OIL image for Qinhuangdao land use classification system is established.(2)The comprehensive data base is designed including NDVI, EVI, DEM, slope chart, aspect chart and topographical feature chart.(3)The general classification of land use is established based on unsupervisedclassification. Sample property sheet is collected from the data base in(2). Decision tree is designed according to C4.5 model and the result is superior to unsupervised classification. It not only takes the advantage of spectrum collecting data by unsupervised classification but also complements the disadvantages of unsupervised classification.