遥感图像分类一直是遥感领域研究的热点问题。目前传统的遥感分类分类方法主要是结合光谱信息进行分类，对于图像的空间信息的利用相对较少，然而事实证明，遥感图像的空间信息是相当丰富的。为了提高遥感数据的利用率，本文采用了结合基于变差函数提取的纹理信息进行遥感图像分类的研究方法。 遥感图像的空间特征包括图像的形状，大小，阴影，纹理，位置和布局，本文主要是针对图像的纹理信息进行研究，有效的利用纹理信息对于遥感图像分类具有很重要的实用价值。文中首先分析了目前国内外关于图像分类方法和图像纹理信息的提取方法的现状。然后对本文所使用的变差函数的基本理论和图像分类的两种方法的基本理论进行详细的说明，并结合具体的实例对变差函数用于纹理提取进行详细介绍，在实验过程中，本文对纹理计算的窗口大小，方向选择等问题上进行详细的探讨。同时，在现有的纹理提取算法的基础上，采用了一种新的纹理计算方法-加权变差函数计算方法。在图像分类方法上，本文选择了两种具有代表性的分类方法，一是传统的分类方法：最大似然法。一是计算机分类方法：BP神经网络分类方法（该分类方法对数据源没有严格的要求）。本文将提取的纹理信息与多光谱图像本身的光谱信息相结合进行分类，旨在提高分类精度。实验证明，结合纹理信息的分类方法，不管是用于传统的分类方法还是用于BP神经网络分类方法，精度都有了明显的提高。同时，在对两种分类方法的比较过程中，也可以看出BP神经网络的分类方法要优于传统的最大似然法分类。本次研究可以得出，结合变差函数提取的纹理信息进行遥感图像的分类可以提高图像的分类精度。

Remote sensing image classification is a hot issue in the field of remote sensing research. At present, the traditional classification of remote sensing classification is combined with spectral information for classification and the image spatial information utilization ratio is low, however,it is proved that spatial information of the remote sensing image is very rich. In order to improve the utilization of remote sensing data, this paper adopts a combination of variogram-based texture information extracted from remote sensing image classification method. Space features of remote sensing image include image shape, size, shadow, texture, location and layout. This paper studies on the image of the texture information and effective use of texture information for the remote sensing image classification has important practical value. First, this paper introduces the status of the image classification methods and the extraction method of image texture information at home and abroad. Next, the paper inruoduces the basic theory of the variogram and two kinds of the image classification methods . The paper discusses the window size of texture, the direction of choice and other issues of detail. Meanwhile, the paper adopts a new method of calculating texture - weighted variogram on the basic of the existing texture extraction algorithm . In image classification methods, the paper chooses two representative classification method, the one is the traditional classification method: maximum likelihood, the other is the computer classification method: BP neural network classification method (the data source is not a strict requirement). In this paper, it is in order to improve the accuracy of the classification to adopt a combination of variogram-based texture information extracted from remote sensing image classification method. The experiments show that combining the classification of texture information, regardless of the classification method used in traditional BP neural network is used for classification accuracy has been improved. At the same time, comparison of two classification process can be seen the classification of BP neural network method is superior to the traditional maximum likelihood classification. This study can be drawn, combined with variogram function information extracted from remote sensing image texture classification can improve the accuracy of image classification.