Incorporating Ancillary Data into a Logical Filter for Classified Satellite Imagery

Abstract

Classified imagery is commonly post‐processed before integration into a geographic information system (GIS). Post‐processing often consists of a rectangular majority filter (e.g., 3 × 3 mode filter) being applied to the classified imagery. An alternate type of filter, the polygon mode or object filter, is also occasionally used in post‐processing. The polygon mode filter is similar to the rectangular mode filter, except that rather than using a kernel of fixed size, the regions are determined a priori. The filter is then applied to the regions as a whole, assigning a whole region to be equal to the dominant land cover in that region. The rectangular and polygon filters both reduce the speckle and polygon count in the resulting data. The objectives of this research were to test a polygon mode filter using agricultural field boundaries that were known a priori and to compare the effects of this filter to the more commonly used rectangular filters. The polygon and rectangular mode filters all increased the land cover accuracy and reduced the polygon count. The overall classification accuracies were improved by the filters: 4% for the 3 × 3 rectangular filter, 5% for the 5 × 5 rectangular filter, and 15% for the polygon filter.