Design of Machine Learning for Limes Classification Based Upon Thai Agricultural Standard No. TAS 27-2017

Abstract

Accurately classifying the limes quality of limes according to established standards is paramount for instilling trust in farmers' trading of agricultural produce. Historically, machinery has been employed to categorize the lime quality, with dual objectives of cost reduction and error mitigation, thereby facilitating the classifying process. Nevertheless, deploying such machinery to classify limes in their fresh produce form, intended for consumer sale, has encountered limitations imposed by the stringent criteria stipulated in Thai Agricultural Standards No. TAS 27-2017, a standard derived from the Codex Standard and widely adopted by numerous countries. Considering these constraints, the presented research aims to enhance the efficiency of limes classification, adhering to the standards. The Machine Learning System is designed to recognize and categorize limes based upon their skin color and defects to achieve this goal. This system employed convolutional neural network (CNN) models in conjunction with logistic regression equations, which are unavailable in the literature. The research findings indicate that this system is proficient in accurately presenting lime images and their corresponding quality classes via a Graphical User Interface on a computer screen, achieving an accuracy rate exceeding 90%. The implications of this research extend to the agricultural sector by augmenting the efficacy of Machine Learning for classifying limes in compliance with Thai Agricultural Standard No. TAS 27-2017. Furthermore, the methodology developed in this study can find applicability in classifying other agricultural products.