Rot Corn Grain Classication by Color and Texture Analysis

Due to the constant increase in corn production and exportation volume, it becomes necessary a digital transformation and tasks automatization in this scenario. The technological artifacts are an important tool for agribusiness companies, once they can improve the controls, agility, efficacy and safety of the procedures. Although automatic operations are frequently used, some activities are still executed manually, such as the rot corn grains classification process. The classification aims to analyze the quality and characteristics of the grains according to a predefined commercial pattern, which are established by the companies that commercialize this cereal. According to the identified characteristics, the negotiation might be affected by price reduction or rejection. Therefore, performing the classification properly is essential to increase the profitability and quality of the corn sold by the companies. Thus, this work aims to introduce a methodology for automatic estimation of percentual range of rot grains by global image analysis of corn grain samples. To be closer to the real process, we diversified the rot percentage inside the images samples by weighing the grains instead of counting it. In this methodology, the images were collected by following a capturing pattern. Next, histograms of color and texture descriptors were computed to characterize the input images. Such histograms were submitted to supervised classifiers to estimate the corn grains percentual range. Several classifiers were tested (minimum distance classifier, SVM, Naive Bayes and Decision Tree) to categorize the corn grains percentual range. In these tests, SVM classifier improved by a boosting technique (AdaBoost) achieved an average accuracy up to 99.97%.