<![CDATA[Automatic classification of remotely sensed digital data is recognised as a robust and efficient method for mapping various land-cover types over a large area. However when more abstract concept such as land-use is required the automatic classification methods cannot be fully useful. This is due to the fact that land-use is related to various landscape factors, and cannot be mapped merely based on its spectral reflectance. This study tried to develop a knowledge-based technique that incorporates textural and terrain information of the image scene into a spectral-based decision making process for land-use labelling. To do so. six reflective hands of Landsat Thematic Mapper (TM) covering Semarang-Ungaran area. Central Java, were used. In addition, all bands were then be filtered using the so-called textural filter, which can accentuate several statistical parameters within a given window. .1 variance parameter was chosen in order to extract heterogeneity within every 7x7 pixels. and the l'ariance values of the whole image dalaset were then stored as a set of texture-filtered bands. Three bands with the lowest 'between-band correlations' were chosen and added to the reflective bands. Based on the nine-layer image dataset, a standard multispectral classification using maximum likelihood algorithm was run. Parallel to this process, a visual interpretation using heads-up digitisation was carried out in order to generate a terrain unit map containing land characteristics relevant to spatial distribution of the land-use in the study area. Finally. the terrain unit map was superimposed with the tentative land-corer map derived from the multispectral classification process. A final land-use map was generated from the nnthisource data integration, controlled by a formalised knowledge about ecological relationship between land-cover. land-use, and land characteristics exist in the field. It was found that the overall accuracy level of the final land-use map is higher as compared to the result generated from six-band classification. However, the use of textural filter also created an 'edge-effect', which shows misclassified pixels alongside the borders of particular land-use categories. The edge-effect also leads to lower accuracy levels for the corresponding land-use categories. In addition, based on the research findings, further research agenda was also set up.]]>
Copyrights © 2003
