Study On Information Extraction Method Of Main Features In Bazhou Xinjiang Area Based On GF-2

The high spatial-resolution satellite images have provided us with accurate and comprehensive information.It has been widely applied in forestry, national defense, monitoring of land and resources, natural disaster monitoring, digital city, etc. GF-2 satellite indicates that china has entered the amish time. This study summarizes the previous studies,and discusses the methods of information extraction of roads,buildings,and vegetation with GF-2 remote sensing image as data sources in XinJiang yuli county.This study’s main research conclusions are as follows:(1)The main method of information extraction of road,building and river is that Frist convolution filtering can enhance the image of low frequency and high frequency detail and edge.Secondly Select the appropriate value for image binarization processing, make the target object more bright, simple and easy to extract;thirdly Mathematical morphology filtering method was used to optimize binary result, filter out insignificance information.Then the image segmentation based on edge detection and based on Full lambda image merging algorithm of the schedule will be several object image segmentation,lastly according to different attributes of the object space, the use of object-oriented information extraction method based on rules, the accurate extraction of target information.(2)In the process of building information extraction, we compared the median filter and Laplasse filter method, the median filtering method has a better effect on the building enhancement, and the accuracy of the building extraction has reached 72.6%.River information extraction method was based on the traditional Laplasse operator to do the appropriate improvement, so that the river information was more prominent, the extraction accuracy reached 86.6%.(3)According to the image characteristics and geometric features of road information in the study area, the road was divided into three categories:County Road, Township Road, simple road. According to different types of road features, different extraction methods were selected, and the extraction accuracy of the county-level roads, towns and villages roads and simple roads were 85.6%,73.5% and 72.4%, respectively.(4) Vegetation information extraction mainly includes vegetation types and vegetation coverage. The vegetation information extraction mainly used the object-oriented information extraction based on the sample,in this study, we compared the results of different segmentation merging scales on image segmentation., and select 50 as the threshold of the Split Merge. In the classification method, we compared the K neighbor method (KNN) with the support vector machine (SVM) classification method.The overall accuracy of KNN was 78.55%, Kappa coefficient is 0.688.SVM the overall classification accuracy is 88.97%, Kappa coefficient is 0.83. The results showed that using this method to extract the information from GF-2 remote sensing image as the data source has higher accuracy and higher reliability.Extraction of vegetation coverage used pixel dichotomy model calculate the vegetation coverage of the study area.