Automatic Segmentation of the Constituent Elements of Bakery Products using the U-Net Architecture

This article explores the efficient application of the U-Net architecture for automatic segmentation of various bakery items. The methodology includes the preparation of a variety of data, including stuffing, dough, and pieces of meat, as well as the application of various augmentation methods, such as horizontal and vertical reflection, elastic transformations, and mesh distortions. The main emphasis is on overcoming the difficulties associated with the similarity of textures, colors and structures between the various constituents of baked goods. Using carefully selected U-Net architecture and a variety of augmentation techniques, it was possible to successfully highlight the contours and boundaries of the filling, dough and pieces of meat in the images. The results of the study demonstrate that the model is able to effectively cope with the segmentation task even in conditions of similarity between objects. This opens up prospects for automating the analysis and classification of ingredients in bakery products, increasing the efficiency of production and quality control. The proposed method can be in demand in the bakery and food industries, contributing to the optimization of production processes and improving the experience of consumers.