Abstract
Present paper deals with instrumentation of image processing and statistical pattern recognition of plant growth in on-line computer system. The nine images of a cucumber plant were taken through the vidicon camera from vertical angle of 45°, by revolving the plants on the turn table from horizontal angle of 0° to 360° at an interval of 40°. A plant image was digitized with binary notation of 256×256 in image processor by setting a constant slice level of the brightness of the plant. The digitized image was read into CPU and stored in disk memories. The nine digitized images were summed to make a synthesized image which was expressed as the compounded matrix of Eq. (2) . Clear difference in growth was observed in numerical increase and distribution of the elements of the matrix. From calculation of sum (P) of the elements, the quantitative growth was represented by the fuction of P as given by Eq. (3) . Further, in an attempt to design the classifiers for estimation of difference in feature of growth, the vertical position (Jl) of centroid in the compounded matrix was calculated by Eqs. (4) and (5), and relation between Jl and P was ana-lyzed statistically : From covariance analysis, regression equation of normal growth was different from that of succulent growth under 29-33°C as shown in Fig. 6. The 95% confidence limits of each regression equation given by Eq. (6) were used as the discriminant function. As shown in Fig. 7, when coordinate of the point (Pi, Jli) belongs to region N, the growth is estimated to be normal. When the point belongs to region S, the growth is estimated to be succulent.
Thus, by image processing of plants, the growth was represented quantitatively as functions of P, and statistical pattern recognition in the coordinate system of P and Jl made it possible to estimate the differences in feature of the growth.
