Unsupervised Classification of Geometrical Data Based on Particle Swarm Optimization
San Ratanasanya, Jumpol Polvichai, Booncharoen Sirinaovakul · 2018
Land use or Land cover management is fundamental process for agricultural development. Remote sensing image can provide useful geometrical data for agricultural development. The unsupervised classification plays an important role to build the map of land use. The accuracy of classification is therefore crucial. There are several classification methods such as K-Mean, ISODATA and Self-Organizing Map (SOM) that are applied to classify land use on geometrical data. However, none of these method can provide the classification accuracy over 52%. The classification method with high accuracy is indeed needed. Particle Swarm Optimization (PSO) could be used to do classification tasks. However, the standard PSO is not suitable to unsupervised classification of remote sensing image. Unlike other complex variation of PSO designed for unsupervised classification, this paper presents the simple adaptation of PSO to unsupervised classification of land use. The sample area is in Ayutthaya province, Thailand which is agricultural area. The field survey is sampled to assess the classification accuracy of the proposed method. The assessment concluded that the proposed method has accuracy up to 73.00% which has approximately 1.4 times more accurate than previous method based on SOM.