Application Research On Remote Sensing Image Classification Based On Geostatistics And ANN

Remote sensing image classification is always a pivotal part of remote sensing study. Most of traditional classification methods are computational auto-classification based on the image spectral characteristics and ignore the spatial textures of image. Furthermore, the traditional classification methods have rigorous limitation to the data distribution. Aim to this problem, this article has put forward to a neural network method on the basis of variogram texture.Texture is the key character of remote sensing image classification and a lot of studies on this have been done. Recently, the appearance of high spatial resolution remote sensing images provides more abundant and clearer texture information, which is the basis of texture classification. This article analyzes the current study situation of remote sensing image classification methods and extracting textural information. Moreover, it analyzes the theory of geostatistics. Based on the geostatistics theory, the variogram is applied to extracting textural information of remote sensing image in this article. It has been proven that the textural information can be used to classification by means of test. At the same time, this article discusses the size of computation window, computation direction and step according to the practical application and puts forward to a auto-adaptive method to determine the size of computation window, In addition, it advances a new method to compute textural information, weighted variogram. Considering that the neural network classification has no limitation to data, this study adopts the back propagation neural network method to classify and recognize the matter combining the textural information extracted by variogram and spectral information. Then the classification results are compared with those gained by maximum likelihood method. The analysis result shows that this method can improve the classification precision. It proves that it is ideal combining spectral features and textural measures based on the geostatistics theory to the classification of the remote sensing image .