Perceptron neural network to evaluate soybean plant shape

In agriculture, human visual judgments take important roles. The visual selection on plant shape in breeding process is one example of such judgments. In this study, in order to develop a stable and generalized plant shape evaluator that can substitute for human visual judgments, we examined perceptron neural network system. We developed a three layers perceptron neural network simulator with direct image inputs. We examined the replacement with such the human visual judgments by the simulator. The matches between the simulator judgments and the human visual judgments, were approximately 60-80%. Though we also examined the relationship between the number of unites and the success rates, we were not able to find any relationship between them. We need to modify the network structure to obtain more appropriate judgments on plant shape