Fuzzy Logic Evaluation of Soybean Plant Shape.

Abstract

Soybean plant shape evaluation is an important part of the soybean plant breeding process in Japan. This selection process is currently performed by visual inspection by the soybean plant breeder. This paper describes a method to evaluate soybean plant shape quality automatically. The method developed was an expert system using fuzzy logic rule sets to evaluate soybean shape quality. The evaluator operated on 4 shape indicators extracted from digitized images of each soybean plant. The evaluator placed the shape of each soybean plant into one of three categories: good (3), fair (2), and poor (1). Only those rated as good were selected by the soybean breeders. The goal was to develop an evaluator that would give the same ratings as those given by the soybean plant breeders. The shape quality evaluation results based on the fuzzy logic rule sets developed in this study were slightly better than those obtained using statistical discriminant analysis. The efficiency of the correct evaluation was about 76% for both the plants to be selected (good) and the plants to be removed (fair or poor). Fuzzy logic evaluation has two advantages in contrast to statistical discriminant analysis. One is that it does not require any assumption on statistical distribution of the shape features and the other is that its structure is easy to understand.