Application Of Multi-layer Information Extraction Approach In The Thematic Mapping From Remotely Sensed Imagery

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 and cover information. It has become an effective means of information extraction of different types in the imagery.To recognize the situation of land use and cover, how to extract various types information from remote sensing image fast and accurately is very important. For the moment, there already have been many classification algorithm, however, owing to the existence of many mixed-pixels and spectrum confusion among different types, it is still a challenge for us to extract land information from medium and low-resolution remotely sensed data.In this paper, Lianshui Basin which is in Xingguo County, Jiangxi province was taken as the study area and the multi-layer information extraction and thematic mapping about land cover based on the images was conducted. First, traditional methods of image classification were explored. Second, the spectrum characteristics of various types were analyzed based on the practical conditions of the study area and the used remotely sensed images, and their divisibility was discussed according to the spectrum features. Third, different combinations of the land cover classification were tested, using the image of Landsat-5 received on December 7th, 1995. During the study, three different classification methods were selected, namely the minimum distance classifier, the decision tree extraction method and the multi-layer information extraction method based on spectrum characteristics and geographic knowledge. Through the analysis and comparison of different methods, the mostly suitable one to the land cover information extraction in the study area was determined.Through this study, the main conclusions are summed up as follows:1) Correctly using the multi-layer classification method, which is based on spectrum characteristics and geographic knowledge, can much improve the classification accuracy of landuse/cover in Lianshui Basin. The minimum distance classifier is a rather good classification method and, however, much attention should be paid to selecting and purification of the training samples.2) With the aid of the digital elevation model (DEM) and its derived information such as slope, aspect, the accuracy of the image classification will be more or less improved.3) The multi-layer information extraction method is more propitious to extrude key point, ensuring the accuracy of key point extraction. This method simplifies complex problems and we can select different classification methods aiming at different objects. During this study, various types of land cover have been extracted successfully on the whole through the techniques of band threshold and the maximum likelihood classifier we selected.4) The multi-layer information extraction method is also propitious to manage the classification results flexibly such as combination of different types, farther partition and so on. By virtue of this advantage, users can expediently divide land cover types they need according to the characteristics of study area and requirement of models.5) The study has shown that the overall accuracy of the classification is 91.79% when using the multi-layer classification method to extract information from TM image. This accuracy can basically meet the demands of the classification and mapping of land cover in the study area.