Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Selection of a Neural Network Architecture for Searching of Pipe Internal Surface Defects O.I. ZAKHAROVA, O.L. KULYAS, A.S. LOSHKAREV, P.A. NAZARENKO, K.A. NIKITIN The object of the study is a computer vision system for objective quality control of the internal surface of cylindrical objects, including pipes for the oil and gas industry. The purpose of the work is to study the capabilities of computer vision methods for identifying defects in various structures of the internal surface of cylindrical objects when using a television system with ultra-wide viewing angles. Neural networks are used as a tool for diagnosing the internal surfaces of cylindrical products, pipes, etc. The parameters of neural network architectures are the size and shape of the convolution kernel, the number of convolutional blocks, the number of subsampling layers, and the number of layers of the output fully connected network. Keywords: computer vision, quality control, surface defect, pipe, television system, image reconstruction, neural network, convolutional network. DOI: 10.25791/pribor.1.2024.1471 Pp. 51-56.