Identification methods of defects in potato tubers to automate the process of their sorting

A method for identifying and separating substandard potato tubers from a common pile based on machine vision and automatic inspection systems is proposed and described. A method based on calculating the color threshold is used for segmenting external defects of potato tubers against the background of a transport conveyor in real time. A centroid tracking algorithm is used to track moving potato tubers. A proprietary dataset consisting of images of commercial and defective potato tubers is created to train the artificial neural network. The results of experimental studies of determining internal defects of potato tubers using nuclear magnetic resonance (NMR) and computed tomography (CT) are presented. A method of controlled impact on a hard surface is used to create hidden defects in the form of darkening of the tuber pulp. The methodology for conducting experimental studies and the operating parameters of NMR and CT are described. A comparative analysis of images obtained using NMR and CT with natural images of tubers in section was carried out, which made it possible to determine with high accuracy the coincidence of the location of defects detected by a non-invasive method with their real location in the tuber. The work demonstrated the value of NMR and CT for a detailed non-invasive method for determining hidden defects of potato tubers on automatic sorting machines.