A comparison of hierarchical and partitional clustering techniques for multispectral image classification

Unsupervised classification of remotely sensed data has traditionally been performed using partitional clustering procedures. This paper compares six classification results for a small Landsat 7 TM sub-image of Hainan Province in China. Of all clustering procedures, the hierarchical nearest neighbour linkage had the lowest classification accuracy, whereas the combinatorial K-means partitional procedure produced the best classification result.