1996 年 58 巻 Supplement 号 p. 497-500
To robotize seedling production, a machine vision system was developed to inspect orchid seedlings and classify into quality and size categories. The boundary of a seedling image was captured with a monochromatic camera and approximated using a polar coordinate. The image features were extracted as Fourier series. A neural network or Bayes decision making were used to develop classifiers for quality and size similar to those used by human inspectors.
Quality coincident-classification rates for poor quality (B) were 92% with the neural network classifier and 89% with the Bayes classifier. Size coincident-classification rates for small size (S), medium size (M) and large size (L) were 82%, 68% and 90% with the neural network classifier and 69%, 73% and 84% with the Bayes classifier respectively.