Technology of Artificial Data Generation for Neural Network Training (using the Example of Quality Control of Laminate Production)

The paper presents the results of a study on artificial data generation methods for computer vision tasks in the operation of conveyor-type technological lines. It identifies the problematic issues of high cost and low efficiency of traditional data collection methods such as video recording and manual markup. A generalized model of synthetic data generation using game engines and 3D modeling is proposed as a solution to these problems. The model was experimentally tested in the context of the laminate coating determination process. Synthetic data allowed the YOLOv8 neural network to be trained with high accuracy, achieving an mAP50 of 0.95. The results indicate that synthetic data can be used to improve the quality of models and optimize learning processes for neural networks in situations where real datasets are limited.