Extracting Vegetation Information Using QuickBird Imagery Based on AdaBoost Classifiers

In this paper, a method called AD-EVI (AdaBoost for Extracting Vegetation Information) based on AdaBoost classifiers is proposed for extracting vegetation information using QuickBird imagery. The proposed approach has the advantages of multi-features selection and combination of weak classifier. The experiment consists of three steps: Firstly, the training samples of vegetation and non-vegetation were selected manually using software ENVI from the given multispectral QuickBird imagery. After the samples selected, the proper feature and correct weak classifiers were chosen for constructing a strong classifier using the training samples. Secondly, the strong classifier was used to distinguish the vegetation from non-vegetation using the QuickBird images. Finally, accuracy and robustness of the AD-EVI method were also verified. It is demonstrated that our method is superior to the compared algorithms (OTSU and minimum distance), and its overall accuracy and Kappa coefficient reach 96.33% and 0.93 respectively. The proposed method can apply in the investigation of land cover and land use for the city management and planning.