Abstract
A neural network system for predicting the mean and variance of normal distribution of characteristics (fruit weight, number of perfect seeds in a fruit, number of imperfect seeds, number of unfertilized seeds, citric acidity and Brix % of fruit juice) of ‘Tosa buntan’ fruit grown on the same tree branch was constructed. Its accuracy level of prediction was evaluated based on the probability concept. As a result, it was found that the probabilities of successful prediction ranged from 77.24% in number of imperfect seeds of fruit to 95.64% in citric acid degree.
Under these probability restrictions of prediction, the optimum cultivating conditions for ‘Tosa buntan’ were estimated by combining the neural network system and the Rosenbrock method as an optimizer. In the calculating process, a proximity which was defined as the product of the target fruit vector and the output vector from the neural network was used as the index for the optimizing level.
The results of the optimization revealed that there are no optimum cultivating conditions which satisfy all the required conditions for the best fruit. Instead, it gave the optimum cultivating conditions for producing the heaviest for fruit, those for obtaining the least seeds in fruit and for optimizing citric acidity, and those for obtaining the maximum Brix % of fruit as a second resolution.
