Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Development of a Neuromorphic Model of a Generative Moment Matching Network for the Synthesis of a Training Dataset of Defects in Rolled Metal Products K.V. MORTIN, А.A. SANDULYAK The article presents the development of a neuromorphic generative model based on the concept of moment matching. This model is designed to synthesize training datasets containing images of rolled metal defects. The proposed approach combines the principles of neuromorphic design, simulating biological neural networks, with deep learning methods to create highly realistic synthetic images of surface defects (cracks, bubbles, corrosion). The model implements a mechanism for matching statistical moments between real and generated data, ensuring their visual and structural similarity. Experiments have shown the effectiveness of the model in increasing the size of the training sample, improving the generalizing ability of neural network algorithms for quality control of rolled metal, as well as in reducing dependence on manual marking of real defective samples. The results of the work can be applied in automatic non-destructive testing systems to increase the reliability and accuracy of detecting defects at early stages of production. Keywords: neuromorphic model, generative network, moment matching, data synthesis, metal defects, small sample learning. DOI: 10.25791/pribor.3.2026.1655 Pp. 01-13.