Experimental Study Of The Knowledge-based Classification Of Remote Sensing Information

Possessing the advantages of macroscopic information, periodicity, real time and. integrity, remote sensing data undoubtedly has already played more and more important role in the gain of Land-use information, and has become an effective technology of extracting features information. It is important to extract all kinds of feature information in remote sensing images quickly and correctly for land-use-and-cover information and land resource planning-and-management. At present, academic circle has put forward many algorithms of computer automatic classification for different types of remote sensing images, but so far it has not established an automatic classification algorithm which has flexibility and high classification accuracy.Expert system is the thinking process of solving and reducing conclusions by simulating human expert's thoughts. This technology has great leap on disposing remote sensing images as opposed to traditional classification which based on pixel statistical analysis. Gradually with the increase of knowledge, the enrichment of different samples and quickly frequency of update, the expert system classification can obtain reliable results, and can realize quick classification and extraction for land use. Based on above, this article has introduced the expert classification system based on knowledge.In this paper, the author has chosen a study region in Chuxiong of Yunnan Province to classify and extract the types of land use.Researched on this paper, the main conclusions are summed up as follows:(1) Based on the knowledge classification which is assisted with the texture and combined with spatial distribution characteristics, time-phase changed character and visual features, classification accuracy is improved obviously. Generally speaking, knowledge-based expert classification system for the overall accuracy and Kappa coefficient of the results is higher than supervised classification, the test improve the extraction of land use classification accuracy. So, it is a recommended method of remote sensing image classification.(2) Residents, the roads, water and dry land has better performance in the extraction, production accuracy has been improved, the production accuracy of dry land was improved as high as 20%. However, there are still leaking classified points and sub-phenomenon mistake.(3) Knowledge-based expert system decision tree classification, will simplify the complexity of the problem, and you can target different definition of the classification by different rules. With the increase of knowledge, it will be more rapid, accurate and reliable to obtain classification results using the knowledge classification. To take full advantage of the knowledge of existing land-use types, such knowledge will be used for remote sensing image classification, it can reduce same object with different spectra, different objects with same spectrum, and improve classification accuracy.(4) Knowledge-based expert system classification method has flexibility in the use of the advantages of various types of information, its efficiency and accuracy depends on the knowledge rule refined or not and the accuracy of the threshold. At the process of building a knowledge base, knowledge rules set up should be given different.