Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Article title: Study of Neural Network Efficiency for Weld Seam Recognition in Images I.A. BOROVITSKY, M.A. LEONTYEV, A.V. YUGAY, A.E. PANFILOV, I.M. KHARITONOV The paper addresses the task of automatic weld seam recognition on images of metallic workpieces using computer vision methods and convolutional neural networks. The study investigates the impact of data preprocessing (image resizing and conversion to grayscale) on the training quality and runtime of the neural network model. The experiments utilized a dataset of more than 8,000 images obtained under real production conditions. The results demonstrate that reducing image size significantly decreases training time while maintaining high recognition accuracy, whereas conversion to grayscale does not notably affect classification quality but increases training time due to additional processing. The findings confirm the feasibility of scaling images to an optimal resolution and using color images when sufficient computational resources are available. Keywords: neural networks, computer vision, weld seam, image recognition, convolutional neural network, data preprocessing, resolution reduction, grayscale, training time, classification accuracy. DOI: 10.25791/pribor.10.2025.1622 Pp. 31-36.