TIPS

Input Representation

One of the Feature Analyst® software’s greatest advantages is its ability to consider both spectral and spatial context when classifying features of interest. Feature Analyst is unique in that it takes into account the area surrounding a target feature instead of relying on spectral signature alone. It is this surrounding information, or spatial context, that is often essential in distinguishing one feature from another. To recognize spatial context, Feature Analyst uses Input Representation.

Generally, it is best to use Bull’s Eye patterns for roads, rivers, and other linear features. In the example shown below, we have selected the Bull’s Eye 2 pattern at a width of 17 pixels.

The Feature Analyst Learner overlays the chosen Input Representation pattern on the image and records the individual color values from all of the pixels under the pattern, as shown in the above animation (click your browser's "Refresh" button to repeat the video).

Feature Analyst applies the Input Representation pattern along the entire set of training polygons that you have already created, collecting data from all of the recorded target pixels. Then, after processing the data, Feature Analyst applies what it has learned from this process, and the Learner extracts features throughout the image that have the same spectral and spatial information.

Click here to view the print version of this Tip of the Week.