Classification Of Remotely Sensed Imagery Based On LVQ Hierarchical Model

The remote sensing technology, which emerged in 1960s and had been promoted by Physics, Informatics, Geography, has following characteristics: (1) Effectiveness. (2) Large scope. (3) The Integration quality of data. As one major technique for people to observe information of the surface of the earth, remote sensing has become more and more important for our lives and our social. Unfortunately, our ability for abstracting information from remotely sensed imagery still largely lags behind technical developments, so it limits the application of remote sensing.The classification of remotely sensed imagery is one of the important research content in process of remotely sensed imagery. It has been widely applied in the research on LULC classification. However, a common problem when classifying remotely sensed imagery in order to map land use/cover is the uncertainty in the process of classification: different land use/cover classes present similar spectral signatures and the same land use/cover classes present dissimilar spectral signatures. Because of these, it is very difficult for traditional spectrum-based classification to meet the demand of precision. Therefore, to improve remotely sensed imagery classification accuracy is one of the main topics in the field of the remote sensing research.In many cases, ancillary characteristics can provide classification with useful information. Also, many attempts have been made to use ancillary characteristics for classification. Ancillary characteristics are incorporated into classification by two approaches: (1) Ancillary characteristics, regarded as logical channels, are input into classification together with spectrum characteristics; (2) The whole image is divided into separate small area based on ancillary characteristics. The two methods above have shortcomings. Compared with other classification algorithms, normal distribution model is not needed for artificial neural networks. In this paper, LVQ neural network was introduced, consequently, a hierarchical classification model based on it was built. The advantages of this model for combining spectral and ancillary characteristics are that it significantly enhances the exploitation of the information. With remotely sensed imagery as the basic data source, this paper used southern margin of Tengle desert as the study area for classification according to the model.The major works and conclusions are summarized as following:1. Three major kinds of supervised classification methods are systematically summarized. This paper focuses on the principle and correlative conceptions about LVQ neural network, and a hierarchical classification model based on it was built.2. Several ancillary characteristics are put forward according to detailed field investigation and knowledge inference. This paper tries to build up the quantitative index of the ancillary characteristics. They were appended to the process of classification.3. The neural network toolbox of MATLAB was used to design and train the program of neural network, under the environment of MATLAB, TM was classified with the aid of the ancillary characteristics.4. In comparison with spectral classification based on MLH, this procedure allowed a statistically significant increase of accuracy of land use/cover classification (from 73.2% to 79.5%). experimental results show that this model can provide reference for land use/cover classification under complex topographical conditions.