Research on Vegetation classification method of combining Decision tree algorithm and maximum likelihood ratio

摘要

Methods of automatic classification are getting more and more Sophisticated, such as Maximum Likelihood Classification Algorithm, which is based on statistics and its' precision for Non - normal distribution data is pretty low. Some other methods including neural network classification, expert system for classification, fuzzy classification, which comes out recently, are either too complex, or only suitable for users who have higher remote sensing and geology knowledge. In view of this problem, this research chose Abaruoergai country as a model district to deal with the vegetation classification methods investigation, where the Kaschin-Beck disease took place frequently. The combined methods were Decision tree algorithm and Maximum Likelihood Classification Algorithm, which are fairly easy to achieve. According to different vegetation types' spectral signatures, spectral knowledge database was built. The whole precision of it was 95.05% and the Kappa coefficient was 0.9016. Both classification methods are easy to use, the combination of them can compensate each other's insufficient, so that enhanced the precision of the classification. The high precision result provide materials for research on relationship between vegetation cover condition and disease incidence rate, and its request to user's specialization is not really high, which is practical and easy to learn for beginners.