On the integration between image classification and geographical knowledge discovery
遥感影像分类与地学知识发现的集成研究

Great complementarity exists between remote sensing image data and geographical knowledge. This paper tries to unify the image classification and geographical knowledge discovery through ground classes data, i.e.,to discover geographical knowledge with remote sensing data drive, to confirm, explain and evaluate image classification result with geographical knowledge, and to represent geographical knowledge with statistic value and distribution atlas. All these come to be an incorporated Remote Sensing and Geographic Classification System. The steps of this method are as follows: firstly, to divide the image into relative big number(>20) of classes using the unsupervised classification; then overlay these unknown classes with the DEM data and get some statistic values and distribution atlas for each class; finally use these values and atlases to name,explain and evaluate each class of the classification result. Meanwhile the correlation between the ground object type and the topographical data is acquired and expressed as well. The example shows that this method makes the classification more efficient and reliable, and it is useful to express and discover the geographical knowledge. The conclusion is that, we can use other data to interpret the result of unsurpervised classification, name and check each class, and at the same time, acquire the geographical knowledge from the pattern in the image data.