Object-based classification for mangrove with VHR remotely sensed image

In remotely sensed imagery with high spatial resolution, more detail spatial information of mangrove forest can be shown. It is important to find a method to effectively use the spatial information so as to improve the accuracy of mangrove forest classification. In the study, different classification schemes (including pixel-based classification and object-based classification), different classifiers, and different texture features have been conducted. The classification results of SPOT-5 image of Matang Mangrove Forest Reserve in Malaysia show that the performances of object-based classifications are better than that of pixel-based classifications. However, the classifier type is important for object-based classification. The accuracies of nearest neighborhood classifiers, which are widely used in object-based classifications, were obviously lower that that of maximum likelihood classifiers and support vector Machines. It is also shown that the involvement of second-order texture features can't effectively improve the classification accuracy of neither object-based classifications nor pixel-based classifications.