Research Progress Report 279 March, 1967
Automatic Identification and Classification of Wheat by Remote Sensing
by David A. Landgrebe and Staff of Laboratory for Agricultural Remote Sensing, Purdue University

Introduction
The primary purpose of this paper is to provide a brief report on early results in the automatic classification of crops from airborne multispectral data by the Purdue Laboratory for Agricultural Remote Sensing.
The specific problem treated here is the automatic detection of wheat in late June.

Remote multispectral sensing research has been carried on at Purdue University for about one year. This research on remote sensing in agriculture is directed at establishing methods to determine by remote means, species identification, state of maturity, disease conditions, soil types, soil moisture conditions and many other crop and soil parameters.

Research workers from engineering and the life sciences are cooperating at Purdue to develop techniques to sense remotely (observe and record from above ground level) agricultural situations and to assess automatically (review and determine by computer) their important characteristics.

Early Results in the Automatic Classification of Wheat
In the automatic detection of wheat in late June reported here, it should be noted that
wheat in the test site region (northcentral Indiana) is near maturity. Results discussed are contained on the four computer printouts which may be interpreted with the aid of the aerial photograph showing the flight line.

Figure 1 shows a portion of a computer printout of the scanner imagery in pictorial form. The computer program which produces printouts of this type was written so that data may be conveniently edited. Column numbers at the top of the printout, and line numbers at the left edge of the printout, are used to determine the address of any given data point. A comparison between the photograph and this printout reveals a number of distinctive field patterns such that orientation from the photograph to the printout is easily possible. For example, consider field A at lines 1381 to 1543 and columns 1 to 115 (Figure 5 presents this area in larger scale), which shows a wheat field with oats planted in the center of it. (Column numbers proceed from right to left due to characteristics of the airborne scanner.)

The results of a first, very unsophisticated or simple, attempt at the automatic classification of wheat is shown on Figure 2 and enlarged in Figure 6. The method used in generating these results was as follows:
In one channel of data, the mean and variance of wheat samples were determined. The

PURDUE UNIVERSITY • Agricultural Experiment Station • Lafayette, Indiana