A refined quadtree-based automatic classification method for remote sensing image

In pixel-based remote sensing image classification, the long processing time limits application of classification. Image segmentation is adopted to accelerate the classification speed. Image segmentation is a procedure of dividing an image into separated homogenous regions. These regions are considered as objects to be classified. A refined quadtree-based segmentation algorithm is proposed in the paper. The windowed aggregation method is designed to solve the problem of over-segmentation, which occurs in quadtree-based segmentation. A spot 5 remote sensing image in Qingdao was selected as the test image. Three experiments were implemented on the test image: the first is pixel-based classification; the second is quadtree-based classification; the third is refined quadtree-based classification. The pixel-based classification obtains the highest accuracy while takes more time. The refined quadtree-based classification is superior to quadtree-based classification in time consumed and accuracy.