摘要：Remote sensing as both a research technique and a research topic is of growing importance to the field of geography. This study is concerned with the development and testing of techniques for digital satellite data processing. To evaluate the suitability of the programs and the classification method, a test was made consisting of a land use classification of the island of Hailuoto (Finland) made on the basis of Landsat 5 TM imagery dated 20-07-1987. The classification method involved the identification of homogeneous training areas for each land use unit from digital data using a colour terminal. The spectral response of these pure areas is used to classify all other areas with a similar spectral response. The final classification is performed using a minimum distance classifier. The results are smoothed with a modal filter and reproduced on a electrostatic colour plotter. The result of the land use classification is shown in Appendix I. The accuracy of the classification is 85 %. Comparison with the forestry and land use classification of the National Board of Survey of Finland showed that the present study is far more accurate. Most errors are due to the confusion of land use classes with their natural counterparts. Aquatic vegetation in shallow water tends to produce a spectral response similar to spruce and birch forest.
Loading...
