Quantification of textures-textural parameters and their significance for classifying agricultural crop types from colour aerial photographs

Abstract

This study demonstrates how aerial photographic textures of agricultural crops can be quantified by deriving measures from small square areas that have been converted to digitized arrays. A Zeiss Scanning-Microscope-Photometer 05 was used as a scanning digitizing instrument and measurements were repeated with light of three different wave-lengths. The output in form of a matrix was processed further by statistical methods to evaluate various types of textural parameters: 1. (1) the standard deviation as value for the overall variation around an average value; 2. (2) the average distance between picture elements belonging to the same density class in row direction and in column direction; 3. (3) the average run-length, i.e., the number of picture elements with values in the same class following each other in row direction and in column direction; 4. (4) the probability that a picture element has a neighbouring point with a value in a particular density class in row direction and in column direction. A discriminant analysis was then employed to determined how many crop fields could be classified correctly with these textural parameters.