Laserfiche WebLink
<br /> <br />. <br /> <br />. <br /> <br />. <br /> <br />Creation of Mask Coverages <br /> <br />o <br />I~) <br />c:) <br />W <br />c.o <br />w <br /> <br />Fields with no crop type information were subset from the vector <br />database. These coverages were used as masks to insure that classified <br />image data were analyzed only in those areas containing fields with no crop <br />information. By eliminating data handling in those areas that were already <br />known, the required analysis was minimized, thereby expediting the work. <br /> <br />Use of TM Data to Determine Crop Type <br /> <br />A major obstacle in using classified TM data to derive crop type <br />information was the spectral variability within a given field. Although the <br />different spectral classes were grouped appropriately in the classification <br />step, how could these classes be analyzed to accurately determine a single <br />crop type for every unknown field? The desired methodology was to use <br />the field boundaries to constrain the image data. For each field lacking a <br />crop type, all corresponding pixels from the classified TM image would be <br />examined. The objective was to establish a majority rule procedure. All <br />classified pixels within a particular field would be tabulated and the field <br />assigned the class with the highest frequency. It should be noted that the <br />class that occurs most frequently also covers the largest area, due to the <br />uniform pixel size (25 meters). While simple in theory, implementation <br />proved challenging. <br /> <br />The initial difficulty was handling image data that were in raster <br />format and field boundaries that were vectors. Numerous complications <br />were encountered in attempting to handle the data within ARC/INFO. After <br />consultation with ARC/INFO experts in Redlands, CA, an AML program was <br />written to assign crop codes to fields based upon spectral information in <br />the classified image. It was immediatelyobvioushowever, that this <br />program was not well-suited to large data sets. The computational <br />inefficiency was staggering and it became apparent that the most efficient <br />approach would be to process the data external to ARC/INFO. <br /> <br />Discussions were held with Paul Weghorst and Darryl Dyke (D-5752) <br />regarding pOSSible options to solve this majority rule problem. After <br />several hours of brainstorming, a technique was developed that would <br />efficiently reformat the data external to ARC/INFO. The end result of this <br />step would be a table containing a unique field-id, the majority class for <br />that field, and the percentage of the field covered by the majority class. <br />This table could then be used to relate the majority class information back <br />to the field polygons contained in the vector database. Numerous <br />programs were developed and linked together to automate this procedure. <br />Modular software development simplified the processing and made error <br /> <br />10 <br />