Segmentation of remote sensing images using multistage unsupervised learning

Abstract

In this study, we investigate an unsupervised learning algorithm for the segmentation of remote sensing images in which the optimum number of clusters is automatically estimated, and the clustering quality is checked. The computational load is also reduced as compared to a single stage algorithm. The algorithm has two stages. At the first stage of the algorithm, the self-organizing map was used to obtain a large number of prototype clusters. At the second stage, these prototype clusters were further clustered with the K-means clustering algorithm to obtain the final clusters. A clustering validity checking method, Davies-Bouldin validity checking index, was used in the second stage of the algorithm to estimate the optimal number of clusters in the data set.

An unsupervised method of classifying remotely sensed images using Kohonen self‐organizing maps and agglomerative hierarchical clustering methods

“…In the literature there are some proposed algorithms that seek to automatically interpret the neurons of a trained SOM (Costa and Netto 2001, Kiang 2001, Wu and Chow 2004, Gonçalves et al 2006. Using this strategy, Sezgin et al (2004) proposed an unsupervised learning algorithm for the classification of remote sensing images that first uses the SOM to obtain a large number of clusters, and then applies the K-means clustering algorithm to automatically segment the trained SOM, and to obtain the final clusters. Although the proposed method has used a clustering validity method to estimate the best number of clusters, approaches that use the K-means algorithm for SOM clustering are only feasible for hyperspherical-shaped clusters (Wu and Chow 2004).…”

“…This index was chosen because it is based on two important concepts, the intra-cluster and inter-cluster density, and also because it is adequate to evaluate data clusters that have arbitrary and complex formats, which does not occur in the majority of other indices. In Sezgin et al (2004), for example, the Davies-Bouldin validation index was used to estimate the best number of clusters in remotely sensed images from the SOM partitioning by the K-means algorithm. This same validation index was also employed in a way combined with another index in the unsupervised classification method proposed in Marçal and Borges (2005).…”

Self-Organizing Map-Based Color Image Segmentation with k-Means Clustering and Saliency Map

“…Rarely simple SOM is implemented directly on image segmentation. Some researchers modified and expanded the typical SOM [11], while others combined SOM with other methods [6,7,10,[12][13][14][15]. Araújo and Costa [11] presented a new SOM with a variable topology for image segmentation.…”

“…Both SOM-K and SOM-KS are guided by an entropy index for image segmentation evaluation to select a best segmentation. To our knowledge, the combination of SOM and k-means was first appeared in [6]. A k-means clustering method with Davies-Bouldin (DB) validity index is implemented on SOM prototype vectors to segment remote sensing images and the results are declared acceptable and efficient.…”

Land-Cover Classification Using Self-Organizing Maps Clustered with Spectral and Spatial Information

“…Land-Cover Classification Using Self-Organizing Maps Clustered with Spectral and Spatial Information 305 hierarchical clustering, are those most commonly applied (Sezgin et al, 2004;Vesanto & Alhoniemi, 2000;Wang, 2002). However, it is worth noting that methods based on K-means algorithm are only feasible for hyper-spherical-shaped clusters and approaches based on classical hierarchical clustering only use inter-cluster distance information to merge the nearest neighboring clusters.…”