Recommendations for choice of interpretive features

By default at creation of project an average color onto all channels and standard deviation onto the second channel (for panchromatic onto the first) is set. These features are enough for recognizing vegetation and hydrography by color and panchromatic pictures. For multispectral pictures it is more preferable to use NDVI channels, especially in a combination with the infrared channel (Nir channel).

The program interface allows to enter a large number of interpretive features. However, we must bear in mind that the increase in features quantity leads to two problems associated with training of the classifier:

- increasing the number of features increases exponentially the volume of demanded memory for storage of a multidimensional file of features distribution;

- increasing the number of features to save the representativeness of training sample it is necessary exponentially to increase its size.

Therefore it is not recommended to increase the number of features more than 4-6 (depending on the size of the training sample - size of patterns).