" : " (WITA-2001)

9-14 2001 , ,



Aircraft photographs and numerical space information are leading source to the map-ping for estimation of the ecosystem diversity. Two ways of the interpretation of aircraft and space information are used. The first is manual interpretation (vector mapping), the second digital treatment (raster images processing). Both the first and the sec-ond approaches, when they are applied to estimation of the ecosystem diversity, have some similar problems.

The manual interpretation of aircraft and space images uses (wittingly or unwittingly) two principles: (1) the map should be sightly and (2) the legend of map cant contain too much of items (ecosystem types). The last principle makes an inability of the com-parison of maps of different areas. In areas with high complexity we tend to combine many items in several and to make complex elements. In monotonous landscapes we may presume to show more faintly discernible items. Thus in the first case the ecosys-tem diversity will be underestimated, but in the second case overestimated. Tendency to make a map beautiful force us to emphasize of distinctive landscape elements and to disguise occasional elements. Thus the results of such mapping are different on different parts of a map. All those make vector maps prepared by manual interpretation very sub-jective.

Raster image processing have similar imperfections but from another source. As with vector mapping, we cant correct compare image classifications on different images be-cause of the lasts have a different time of scanning, different atmosphere condition and different landscape pattern. The landscape differences lead to different training sets, which are using in clastering procedures. Thus the algorithm elaborated to classification of the one image cant be used on the other image. Raster image processing gives a uni-form result on whole area of image. It is the advantage of manual interpretation but it is slightly darkled in compare with low quality of digital image classification. Formal al-gorithm does not use the contextual information, which is used under manual interpreta-tion of remote sensing data. Quality of results also determined by using textural infor-mation. Raster images processing also contain subjective elements: selection of training set, classify procedure and other, which are permit high variableness of results of classi-fication.

The superiority of digital processing before manual interpretation is high speed of re-sults obtaining but improvement of results quality increase cost and time of treatment. The quality of image interpretation is advantage of manual treatment but this way of mapping is very slowly. To increase of speed of manual interpretation we need to enlarge number of working personal, but it will increase of cost and decrease of quality, because of high subjectivity of process. Selection between two alternative approaches is determined as balance between cost, quality and speed.

Work was done in framework of INTAS-99-01718 project and N.66 SB RAS integra-tion project.


[ ]

1996-2000, ,
: 06-Jul-2012 (11:44:54)