An improved random decision trees algorithm with application to land cover classification

Random decision trees algorithm, which is no need to count images and has good portability, is a great prospective method for land cover remote sensing classification at present. An improved random decision trees algorithm with application to land cover remote sensing classification was proposed in this paper. Firstly, in accordance with the low operation efficiency of random decision trees algorithm, an improved random decision trees algorithm was presented by adding tree balance factor, setting node impurity and distinguishing sample types. Secondly, by taking the ALOS images of Longmen city of Guangdong Province in China as study object, the remote sensing classification was conducted using the improved random decision trees algorithm. Finally, a comparison study was proceeded to compare the improved random decision trees algorithm with maximum likelihood classification method. The results indicate that the classification precision is improved from 81.46% to 87.53% and Kappa coefficient is up to 0.8524. By taking extreme imbalanced decision trees, node impurity and distinguishing sample types into consideration, the improved random decision trees algorithm can improve the efficiency and accuracy of land cover remote sensing classification.