A Study On Multi-spectral Remote Sensing Digital Image Classification

The development of remote sensing has offered abundant observing the data to mankind. And a trend of quantitative analysis of remotely sensed information is gradually realizing information extracted from the image automatically. It is not only the demand of the remote sensing application but also the advance of the remote sensing self-development. There is a very important demand at present in automatic classification of topographical objects by utilizing the remotely sensed imagery. However, automatic classification based on computer systems is still a challenge field because of the limitations of the complexity of natural conditions and remote sensing technology itself.The aim of this paper is to study classification approach of topographical objects with Multi-spectral image. In this thesis, a hierarchical classification and interpretation model is presented, which consists of four parts: a target model, original regions detection, hierarchical classification and class interpretation. The target model is used to search class characters based on analysis of training data. In the process of detecting original regions, a best wave band is selected based on the target model, which has good separability and can used to segment the images, then original regions are generated Subsequently, in the part of the hierarchical classification, a structure of hierarchical extracting topographical objects is set up. Finally, in the part of the class interpretation part, each region classified is labeled into one of object categories. Relying on the hierarchical classification and interpretation model, a grading hierarchical classification algorithm is develpoed, then applied to the classification of topographical objects in the partial region of Daqing. The experimental results show the classification acuuracy by our approach are obviousely higher than the results by the traditional approaches.