The Application of Classification and Regression Tree (CART) with Multi-feature Image on Land Cover Classification in Mining Area

摘要

In the mining area, the situation of land cover is usually quite complicated because of the mining activity. The accuracy of land cover classification plays a significant role in the rational exploitation of mineral resources and effective monitoring of the ecological environment in mining areas. Nowadays, incorporating the spectral characteristic, texture features and multi-source auxiliary data into image classification can increase classification accuracy and precision, but a basic problem is that using the conventional classification methods is too hard to find hidden and useful information in mass multifeature data. In this study, we discuss Classification and Regression Tree (CART), a data mining algorithm, and use it into the land cover classification in Huangshi mining area, taking multicharacteristic data composed with remote sensing image, feature indexes, texture information as input variables. And we obtain 97 classification rules. According to classification evaluation, the CART algorithm improve the classification accuracy by 3.98%, 10.26% respectively compared with supervised classification using Minimum Distance and unsupervised classification using Iterative SelfOrganizing Data Analysis Technique (ISODATA), and the Kappa coefficient increases by 0.0733, 0.1301 respectively, which can provide a strong support for interpretation of the geological environment for the mining areas.