A Study on the Accurate Classification Approaches for Remote Sensing Image Based on GIS
GIS支持下的湿地遥感信息高精度分类方法研究

How to get high accuracy of classification by remote sensing image processing technology is oneof the difficulties in applications of wetland data. The paper discusses the methods to enhance the accuracyof classification for the wetland of Zhalong. First we can noise reductions enhance and tasseled cap enhanceto improve its accuracy. Then compare the classification results of original image with those of the othertwo enhance methods. The results are very similar, i.e., 83.91% of original image, 78.88% of noisereduction and 90.13% of tasseled cap respectively, checked by GPS sample points gained from fields, landcover data of 2000 and Kappa accuracy assessment. Each method has its own advantage in differentiatingthe classes, especially the accuracy in the wetland boundary after noise reduction enhance and the accuracyin the sub classes of wetland after tasseled cap enhance. But none of them is very satisfied. So based onknowledge and the technique of GIS, we can put the advantages of each methods together to gain anexcellent result. The new method is the kernel of this paper, using spatial algebra, which combines thetechnique of RS and GIS together, and it can greatly improve the accuracy and reflect the actual land typesbetter in classification of wetland. Its accuracy can get to 96% after checked by GPS sample points,landcover data of 2000 and visual interpretation. Then we can draw a conclusion that as far as wetlandclassification is concerted, the image enhance before classification can improve its accuracy in some parts,but using spatial algebra based on GIS technology can put the advantage together and get the best result.