Clustering the tropical wood species using Kohonen Self-Organizing Map (KSOM)

Abstract

This paper discusses on how the Kohonen Self Organizing Map (KSOM) is used as a tool to cluster and classify the tropical wood species. Wood features have been extracted through the use of two features extractors; Basic Grey Level Aura Matrix (BGLAM) and Statistical Properties of Pores Distribution (SPPD) techniques from the wood images. The wood dataset is trained and tested separately using KSOM algorithm with different parameters such as the number of epochs and map sizes in order to find the best topological network for clustering and classifying the wood data. The clustering results are analyzed and the best result is selected based on common KSOM performance measurement; topological error and quantization error. The number of cluster performed by KSOM is 61 clusters, while the number of overlapped cluster varies for each map. From the results, the 23x23 map size has produced the lowest number of overlapped clusters with the minimum value of topological error and quantization error.