Optimizing features based on object-oriented classification of a high-resolution urban image from the ZY-3 satellite

Abstract

There is a growing tendency for the utilization of high-resolution images due to the development of the sensors. However, the object-oriented classification cannot meet the accuracy requirement of the high-resolution images due to a large number of irrelevant and unnecessary features. In this dissertation, Yinchuan has been selected as the study area to explore the result of classification of optimized features (11 features) and initial features (41 features) based in object-oriented method on ZY-3 satellite. Correlated-based Feature Selection have been used to optimized features to reduce the abundant and irrelevant features. As a result, the classification of overall accuracy and kappa coefficient of optimized features have improved 6.44% and 8.45% respectively. In addition, the visual effect and processing time of optimized features have also been enhanced to a different extent. Therefore, the combination of feature selection and object-oriented method is better to enhance the classification accuracy of the high-resolution images.