Algorithm of Digital Identification of Grain Materials

Abstract

The article examines the application of machine vision technologies to enhance the accuracy and efficiency of grain material identification during post-harvest processing. It has been determined that traditional grain quality control methods, including visual, microbiological, and chemical tests, have significant limitations in identifying impurities, especially those similar in physical and visual characteristics. In response to these challenges, a digital grain identification algorithm based on machine vision and machine learning methods is proposed.

The algorithm allows for rapid and accurate analysis of grain material images, enabling automatic recognition, classification, and quality assessment. Additionally, the algorithm shows potential for scalability and integration into modern agro-industrial processes, helping to minimize grain loss and improve preservation during transportation and storage. This positions machine vision technologies as a promising tool for ensuring food security in Ukraine. The article summarizes global experience in digital object identification in images and highlights possible methods for identifying grain materials, facilitating the development of the algorithm. A preliminary version of the digital identification algorithm for grain materials has been developed, incorporating relevant methods that enable recognition, classification, and evaluation of grain material indicators.

Convolutional Neural Networks (CNNs) have been found to be the main type of neural networks for working with images, which are specialized for processing data with mesh topology. They have become very popular in computer vision tasks, including object recognition, image segmentation, and classification, due to their ability to automatically train and extract features without the need to explicitly define them manually.

Deep neural networks are now the dominant classification approach because they automatically extract complex features and have high accuracy. The choice of a specific method depends on the availability of data, resources and requirements for classification accuracy.

After classification, it is necessary to determine their specific parameters for objects classified as grains - the fourth stage of the algorithm. These parameters mean the shape of the grain and its geometric dimensions (width and length). The shape of the grain contour reflects the actual shape of the grain.