Abstract: It is a new approach to improve the accuracy of image classification in combining spectral feature with texture and structural features of ground objective on satellite image. Based on the recognizable characteristics of satellite image, it is introduced how to describe and capture texture and structural features of ground objective by the Discrete Fractional Brownian Motion model. Furthermore, neural networks are used for classification tool of satellite image. In classification spectral feature, texture and structural features of ground objective are used for the category of an IRS-1C satellite image. The category result shows this approach is better than the maximum likelihood classifier.

Key words: satellite image classification, fractal, neural networks, texture and structural characteristics

摘要： 根据卫星数字图像特点，引入了分形方法来描述纹理结构特征，利用离散分形布朗运动(DFBM)统计模型来抽取卫星图像纹理结构特征。在此基础上，采用神经网络方法将纹理结构特征与地物光谱特征相结合，进行卫星图像分类。试验结果表明，该分法分类效果优于单纯采用光谱特征分类的最大似然法。

关键词: 卫星数字图像, 分形, 神经网络, 纹理结构特征