A novel hierarchical clustering technique based on splitting and merging

Amongst the multiple benefits and uses of remote sensing, one of the most important applications is to solve the problem of land-cover mapping. In this paper, unsupervised techniques are considered for land-cover mapping using multispectral satellite images. In unsupervised techniques, automatic generation of the number of clusters for huge databases has not been exploited to its full potential. To overcome that, a hierarchical clustering algorithm that uses splitting and merging techniques is proposed here. In the proposed method, a splitting method is initially used to search for the best possible number of clusters with a non-parametric estimation technique, i.e., mean shift clustering (MSC). For the obtained clusters, a merging method is used to group the data points based on a parametric method (k-means clustering algorithm). The performance of the proposed hierarchical clustering algorithm is compared with three previously proposed unsupervised algorithms, i.e., (1) parametric k-means clustering; (2) hybrid MSC and k-means clustering; (3) hybrid algorithm for cluster establishment (ACE) and k-means clustering. Two typical multispectral satellite images – a Landsat 7 thematic mapper image and a QuickBird image are used to demonstrate the performance of the proposed hierarchical clustering algorithm. A performance comparison of this proposed algorithm with the previously proposed algorithms is presented. From the obtained results, it is concluded that the proposed hierarchical clustering algorithm is both more accurate and more robust than the other compared algorithms.