DECISION TREE BASED CLASSIFICATION OF REMOTELY SENSED DATA

摘要

Over the last decade, many applications of neural network classifiers for land use classification have been reported in the literature, but few studies have assessed the use of decision tree classifiers. These latter techniques have substantial advantages for remote sensing classification problems due to their nonparametric nature, simplicity, robustness with respect to non-linear and noisy relations among input features and class labels, and their computational efficiency. A decision tree classifier has a simple form which can be compactly stored and that efficiently classifies new data. Decision tree classifiers can perform automatic feature selection and complexity reduction, while the tree structure gives easily understandable and interpretable information regarding the predictive or generalisation ability of the data. A decision tree recursively partitions a data set into smaller subdivisions on the basis of tests applied to one or more features at each node of the tree. In this study, a decision tree classifier is used for land use classification using Landsat-7 ETM+ data for an agricultural area near Littleport (Cambridgeshire), UK, for the year 2000. A field study was carried out to collect ground truth information about the various land use classes in the study area. Six land use classes (wheat, sugar beet, potatoes, peas, onions and lettuce) are selected for classification, and a univariate decision tree classifier is used for the labelling of the image pixels. The results of this study suggest that the decision tree classifier performs well, producing an overall accuracy of about 84.5%. The boosting technique, which improves the classification accuracy of a base classifier, was applied and the classification accuracy was increased by about 2 percent to 86.5%.