Identification of rice using symbolic regression

Abstract

Image classification is a preferred method for agriculture product identification since the product is intact from the process. Many published works have adopted image classification techniques for identifying rice seed varieties. Based on a classification algorithm and given sample features vectors, a classification model is built from a set of optimal parameters and operators. However, the obtained parameters rarely reflect relations among the feature symbolically and, therefore, are meaningless to human. This paper proposes of using a symbolic regression algorithm to search for possible identification solutions. A number of possible solutions and the associated analytical expressions are obtained. As a case study two classifiers are built from two chosen solutions and are applied to identify high quality Jasmine rice, the Khao Dawk Mali 105, from other three alike varieties. The performance of the classifiers is compared with the published work using percent identification accuracy. The experimental results show that the two classifiers obtained from a symbolic regression algorithm accurately identify the Khao Dawk Mali 105 at 86.25% and 90.00% while the compared algorithm accurately identify the rice in the range of 73.85%-83.46%. The most advantage of using a symbolic regression algorithm is that the algorithm also reveals analytical expressions. Such expressions suggest human a potential methodology concealed in the observed data.