Improvement of classification accuracy in remote sensing using morphological filter

In this paper, we further develop a new pixel-based multispectral image classification algorithm. Our method consists of two stages. First, we use a noise reduction filter using mathematical morphology, and next we use a classification algorithm such as the maximum likelihood classification method with the filtered image. With the new method, we got much better results in terms of both training and testing data accuracies than many other classification algorithms like minimum Euclidean distance, Fisher linear likelihood and extraction and classification of homogeneous objects, which is a spectral-spatial classifier algorithm. The thematic maps obtained with the proposed algorithm are also more smooth and acceptable than the others.