The purpose of this work is to study the applicability of the U-Net architecture for automatically determining the contours of natural skins using the TensorFlow and Keras libraries in Python. A software application has been developed based on methods including OpenCV libraries, as well as a model for implementing a deep convolutional neural network. The dataset for training and testing the network was created using augmentation. Training was carried out using the stochastic gradient descent method after splitting the data sample into training and test images. In the future, the results obtained will be used to create an automated system that will make it possible to determine the contours of the skin and its defects, which in turn will open up the possibility of calculating the useful area of ​​the skin and creating an automated layout of patterns taking into account the identified defects.

Keywords: computer vision, edge detection, natural skin, machine learning, convolutional neural networks, U-Net architecture, deep learning

The article discusses correlation methods of image identification. An algorithm of the "rare grid" method has been developed.

Keywords: image identification, algorithm, recognition, cutting, reference frame, element correlations, minimum search

The article discusses the method of recognizing contours in the primary image. An algorithm has been developed for finding the absolute minimum of functionality in an image.

Keywords: contour, algorithm, defect, recognition, cut, reference frame, minimum search