A.Yu. Shafigullina1, D.I. Klimov2, T.T. Mamedov3, I.R. Gubaidullin4

1–4 Joint Stock Company "Russian Space Systems" (Moscow, Russia)

2 National Research University "Moscow Power Engineering Institute" (Moscow, Russia)

1 postgraduate contact@spacecorp.ru, 2 klimov.di@spacecorp.ru,
3 mamedov.tt@spacecorp.ru, 4 gubaidullin.ir@spacecorp.ru

Problem setting. The growing trend in the development of artificial intelligence and in particular machine vision makes it possible to introduce it into various areas of human activity, in particular, into flaw detection of integrated circuits. To date, there are a large number of methods for implementing computer vision, each of which has its pros and cons, and for each task, the result of applying these methods will be different.

Target. The purpose of the work is to review the existing methods of machine vision and experimental studies to determine the most appropriate algorithm for the implementation of the classification of defects in integrated circuits.

Results. As a result of experimental studies within the framework of the problem of classifying defects in integrated circuits, the problematic issues of each classification method were identified, and recommendations were given for more efficient application of the methods.

Practical significance. The authors showed that the best classification accuracy is given by a convolutional neural network (98%), classical training by the k-nearest neighbors method with the condition of identifying image features using scale-invariant transformation (SIFT) gives an accuracy higher than neural networks (50%).

Shafigullina A.Yu., Klimov D.I., Mamedov T.T., Gubaidullin I.R. Modern methods of computer vision and their practical application in the problem of defectoscopy of integrated circuits. Neurocomputers. 2024. V. 26. № 2. Р. 23-33. DOI: https://doi.org/10.18127/ j19998554-202402-03 (In Russian)

1. Adamova A.A., Zaykin V.A., Gordeev D.V. Methods and technologies of machine learning in neural network for computer vision purposes. Neurocomputers. 2021. V. 23. № 4. Р. 25−39. DOI: https://doi.org/10.18127/j19998554-202104-03. (in Russian)
2. Korchagin S.A. Computer vision system for automatic classification of nanocomposites. Information measuring and control systems. 2023. V. 21. № 5. P. 16–26. DOI 10.18127/j20700814-202305-03. (in Russian)
3. Viryasova A.Yu., Ishin I.A. Analysis of the subject area and development of a database for the implementation of flaw detection of integral structures. Academy. 2017. № 12(27). P. 26–29. (in Russian)
4. Tareen S.A.K., Saleem Z. A comparative analysis of SIFT, SURF, KAZE, AKAZE, ORB, and BRISK. International Conference on Computing, Mathematics and Engineering Technologies. Sukkur, Pakistan. 2018. P. 1–10. DOI 10.1109/ICOMET.2018.8346440.
5. Zivkovic S. Feature Matching methods comparison in OpenCV. [Electronic resource] – Access mode: https://datahacker.rs/feature-matching-methods-comparison-in-opencv/, date of reference 30.03.2023.
6. Blei D.M., Ng A.Y., Jordan M.I. Latent Dirichlet Allocation. Journal of Machine Learning Research. 2003. V. 3. P. 993–1022.
7. Nefedov A. Support Vector Machines: A Simple Tutorial. 2016.
8. Ng A.Y., Jordan, M.I. On Discriminative vs. Generative Classifiers: A comparison of Logistic Regression and Naive Bayes. Advances in Neural Information Processing Systems. 2002.