Extracting Fly Ash Site Information Using Decision Tree Classification

Fly ash not only pollutes the environment but also endangers the human health. Rapid, real-time, accurate identification of fly ash by means of remote sensing is of great significance for protecting the environment and the human health. In this paper, by analyzing the spectral information of the typical surface features in Baotou City, based on Landsat 5 TM image data, it adopted the decision tree classification to extract fly ash in the study area. Firstly, we analyze the spectral characteristics of the typical objects and the relationship between them in the study area. Secondly, we established the decision tree, used Soil-adjusted Vegetation Index (SAVI), Modified Normalized Difference Water Index (MNDWI), Normalized Difference Built-up Index (NDBI) and Spectrum Threshold Method to classify the image respectively. Ultimately, post-process the classified image with shape feature and location feature. The total classification accuracy was up to 70.7%. The experimental results show that the method is suitable for the automatic extraction of fly ash information, that what combined with the visual interpretation, can achieve high accuracy.