The aim of the study was to develop an application-programming interface for automatic recognition of images containing massive volumes of sugar beet and quality determination. The REST interface was designed to be integrated into external systems for assessing sugar beet quality from photos of lorries. A neural network model was trained using image partitioning to identify five defects: tops, chips, dirt, frostbite, and snow. In the work, it was important to solve the problem of occlusion that occurs due to the specifics of images in the dataset. To solve the problem of occlusion, a multi-layered neural network structure was used. The processed results generate a quality report for logistics decisions at sugar beet processing plants. The article discusses the main problems of creating this specific neural network. The trained model achieves an average accuracy of 0.91 in recognizing defects in large sugar beet images, enabling automation and acceleration of quality sorting.