Applied Physics and Mathematics Annotation << Back SEGMENTATION OF OXIDE LAYER DEFECTS IN TANTALUM CAPACITORS WITH CONVOLUTIONAL NEURAL NETWORK I.A. ELKIN, A.I. CHUKAVIN, O.B. BARYSHEV The dataset has been prepared and convolutional neural network has been trained. The resulting model makes it possible to segment defects in the dielectric layer of tantalum metal oxide capacitors after electrochemical oxidation or after the removal of the cathode layers with high accuracy. It was shown that if we use the model in the sliding window mode and vary the scanning step, we can improve the quality of defects segmentation and analyze large images. Keywords: tantalum capacitors, defects, deep learning, semantic segmentation. DOI: 10.25791/pfim.04.2023.1271 Pp. 65-72.