The article proposes an image recognition algorithm for an automated fiberglass defect inspection system using machine learning methods. Various types of neural network architectures are considered, such as models with a firing rate of neurons, a Hopfield network, a restricted Boltzmann machine, and convolutional neural networks. A convolutional neural network of the ResNet model was chosen for developing the algorithm. To develop the algorithm, a convolutional neural network of the ResNet model was selected. As a result of testing the program, the neural network worked correctly with a high learning percentage.

Keywords: fiberglass, defects, machine learning, convolutional neural networks, ResNet architecture, testing, accuracy