The Research On Remote Sensing Image Classification Based On Fractal Theoretics

Remote sensing, which came into being in 1960s and experienced rapid development, is a comprehensive detection technology. The unequalled superiority of remote sensing has been demonstrated by its applications in city layout, resource perambulation, environment protecting, land monitoring, agriculture, forestry and military, and its scope of application keeps expanding. The remote sensing image classification is an important issue in the application of remote sensing technology. Its goal is to classify the pixels of remote sensing image according to terrain feature category.At present, remote sensing image classification method only uses spectrum information, which results in low classification accuracy. Thus it's necessary to bring the texture feature into classification. On the basis of summarizing and comparing the popular texture analysis algorithms, we focus on the remote sensing classification algorithm based on fractal theoretic. Using color differentiation and S component as the color features and fractal dimension as the texture feature, this algorithm classify pixels by combining these two features.The contents of this paper are as follows:1. For most algorithms only use color information, a remote sensing classification algorithm based on fractal theoretic is proposed. It uses differential box counting approach to extract texture feature. The color features are extracted by color differentiation and S component in HSI color model. Then it combines the color features and texture feature as the basis, and uses FCM algorithm to classify. This algorithm improves the accuracy while not reducing the efficiency.2. A remote sensing classification algorithm based on multi-fractal dimensions features is proposed. It not only extracts fractal dimension of the gray-level image, but also calculates other fractal dimensions of high and low gray-level image and gradient image. This algorithm can further improve the classification accuracy by classifying with combined color features.