Methods And Applications Research Of Land Use Spatial Data Mining Based On Wavelet Analysis

The high-resolution remote sensing image is one important component of the land-use spatial data. The basic starting point of this paper is seeking for the more effective methods to get more useful information about land use from the image data. Wavelet analysis is developed in the past two decades, and has been applied in many fields by virtue of its excellent features. How to take advantage of the characteristics of wavelet analysis in land-use spatial data mining is the main content of this paper.In this paper, wavelet analysis was utilized in texture feature extraction of land-use high-resolution remote sensing image, multi-scale information extraction and so on. The major conclusions are as following:1. The two major characteristics of wavelet analysis are the base for its application in land-use spatial data mining. One is the multi-scale decomposition of wavelet analysis, that is, the characteristics of multi-resolution analysis. The other is that texture features is separated forward to the high-frequency direction with the characteristics of direction after the wavelet decomposition of image.2. We got the distribution of energy after decomposing every land use type in land-use remote sensing image, and different land use types perform different characteristics on different scales and directions.(1) The values of texture energy are concentration and interwoven relatively on the first and second scales where it is difficult to carry out all over the extraction of ground material. The separation of the values on third and fourth scales is better, so it is easier to extract the ground material.(2) Images in the wavelet decomposition of the high-frequency components have three directions: horizontal direction, vertical direction and diagonal direction. It is easier to extract ground material in the first two directions than the other direction.3. The difference of the texture energy between different land use types and different bands of remote sensing image is very small, which is almost in a state of overlapping. We can use singe-wave approximation in place of the whole image for extracting the wavelet energy texture feature.4. Form the distribution of wavelet energy on every scale, we can see that the noise energy is concentrated on the first three decomposing scales and the de-noising for wavelet analysis can be carried out on the three scales.