Classification for Multispectral Image Using Spatial Information

A new classification scheme for remotely sensed image is proposed in this paper. First, the mixed pixels are processed by means of the spatial structure information in the image, which improves the recogition accuracy of mixed pixels and increases the separability of low resolution image. Then the processed image is classified using the optimized binary tree classifier. The relation between class's prior probability and computing time is considered when the classifier is designed, which improve classification accuracy and reduce computing time. The results of appliying this scheme to both experimental image and actual Landsat image show excellent.