Sampling Strategy for Multi Spectral Image Classification Using 3-D-DWT and Morphological Profile

Abstract

Nowadays, Spectral-spatial processing has been used in multispectral image classification. In general, we select the training and testing samples randomly from the image. This method leads to overlap between the training and testing samples i.e the training and testing samples are sometimes same. This will give overlap between training and testing samples. To prevent this problem, we propose a controlled random sampling strategy for spectral-spatial methods. It can substantially reduce the overlap between training and testing samples and gives accurate evaluation. The overlap between the training and testing samples enriched by spatial information methods such as morphological profile and spatial filtering. The random sampling method is not right to access the spectral-spatial classification algorithms, since it is difficult to find the improvement of classification accuracy is originated by increasing the overlap between the training and testing samples.