Classificação de grãos de milho da espécie Zea mays da classe amarela utilizando processamento digital de imagens

Abstract

Corn is one of the most produced cereals in Brazil and the world, playing a vital role in the national economy in terms of both domestic consumption and exportation. The qualitative classification process of the grains must be as effective as possible since a significant portion of the cereal production is designated for animal and human consumption. Currently, the classification process is conducted manually by experts, leading to various biases and problems that may arise. Computational processing techniques have been widely used to address everyday problems, among which Digital Image Processing (DIP) techniques stand out in pattern recognition. Thus, this work proposes a classification method primarily focused on the coloration of the grains. Using a device equipped with an ESP-32 CAM, images of samples varying both in grain quantity and defect percentage are captured. After analyzing these images, classification models were generated. The model aimed at estimating the quantity of grains achieved approximately 84% accuracy. The detailed evaluation of the defect percentage reached about 63% accuracy, while the binary classification (defective or healthy grains) presented an accuracy of approximately 88%.