| With the rapid development of remote sensing (RS) technology,there is a far-ranging use of remote sensing image in the field of land use/land cover research .How to improve the accuracy of RS interpretation automatically is a challenge in RS application. On the other hand, the existing geographical data contain plenty of spatial and attribute information. How to apply geographical data into RS classification to improve the accuracy is one of the matters that the scholar pays attention to. It is popularly considered that considering the influence of the prior probability accords with Bayes rule and ensures the minimal loss in the classifying progress. Supported by the analysis and advance process to the geographical data using GIS software, the paper discusses the question that whether the accuracy of Bayes supervised classification will be improved considering the influence of the prior probability.In this paper,using ArcView,in the process of land cover classification of Kaifeng city,the geographical data such as the land use map are made into vector maps, then the assistant data, containing attribute of each land cover sort, are gained. The proportion based on the assistant data is used as the prior probability to replace the prior value in the conventional supervised classification; the farther iterative prior probability is applied into classifying progress on Landsat TM image. The overall accuracy and kappa index are calculated by using the error matrix to check up the classification's accuracy.The main conclusions are following: (1) Compared with the conventional MLC, the method of iterative prior probability based on the vector map can dispel the prior probability's influence and the overall accuracy and kappa index can be improved; (2) To the types with greater area than average area of all types, the producer's accuracy will be improved while user's accuracy be lessened, but to the ones with smaller area, the situation is just the opposite.
|
PDF Full Text Request