Polygon-based image analysis within an integrated GIS/RS environment

Remote sensing (RS) is a promising technique for updating the thematic data about terrain objects stored in a geographic information system (GIS). This can be achieved by means of automatically processing remotely sensed data. Most current automatic techniques operate on a pixel-by-pixel basis in isolation from other pertinent information. Therefore, per-pixel techniques often yield results with limited reliability. The reliability of information extraction from RS can be improved by incorporating apriori knowledge about the thematic information searched for and performing the analysis in a polygon specific manner.; In this research, two polygon based image analysis techniques were developed and successfully implemented within RS and GIS integration. The first technique detects the changes and prevailing land cover conditions within land use boundaries stored in a GIS by digital image analysis. In this technique, the land cover types within land use boundaries to be monitored are assumed known. It consists of two schemes. In the first scheme, a standard pixel-by-pixel image classification is performed, one theme at a time, covering the entire area of interest. A polygon-by-polygon assessment of the results follows to separate those polygons which deviate from the apriori-expectations by more than a preset threshold level. These polygons are then subjected to further examination. In the second scheme, image classification is by-passed entirely and only image statistics are generated polygon-by-polygon. A comparison of these statistics with those representing normal, expected conditions identifies those polygons where significant changes occurred. Both schemes were tested on the monitoring of tree plantations in North-Central New Brunswick.; In certain land units, such as agricultural lands, the land cover is dynamic and must be identified first to perform monitoring operations. The second polygon based image analysis technique was developed therefore to identify the land cover types within land use boundaries stored in a GIS. In this technique, first, the remotely sensed data are classified by a multistage parallelepiped classification logic. Then, a polygon-by-polygon assessment follows where the percentage of classified pixels are computed for each class in each polygon. After that, the results are adjusted by distributing those percentage values falling below a pre-defined threshold (T) limit proportionally to the size of classes having percentage values above T. The technique was tested on an agricultural region located near Woodstock, New Brunswick.; The above polygon based image analysis techniques of monitoring changes and identifying land cover types overcome the problems of multi-classification and misclassification due to overlaps of spectral signatures, which often plague traditional image classification. In addition, the techniques provide for an immediate update of the database in the GIS.