ntvНаучно-технический вестник информационных технологий, механики и оптикиScientific and Technical Journal of Information Technologies, Mechanics and Optics2226-14942500-0373Университет ИТМО10.17586/2226-1494-2025-25-2-273-285ntv-447Research ArticleКОМПЬЮТЕРНЫЕ СИСТЕМЫ И ИНФОРМАЦИОННЫЕ ТЕХНОЛОГИИCOMPUTER SCIENCEСтруктурный анализ изображения и адаптивный резонанс в искусственных нейронных сетях (обзорная статья)Image structural analysis and adaptive resonance in artificial neural networks (Review paper)https://orcid.org/0000-0003-1160-3758ЛуцивВ. Р.LutsivV. R.

Луцив Вадим Ростиславович — доктор технических наук, профессор.

Санкт-Петербург, 190000, sc 6602625465

Vadim R. Lutsiv — D.Sc., Professor.

Saint Petersburg, 190000, sc 6602625465

vluciv@mail.ruhttps://orcid.org/0000-0002-7186-5744МихальковаМ. А.MikhalkovaM. A.

Михалькова Мария Анатольевна — младший научный сотрудник.

Санкт-Петербург, 199034, sc 57218284288

Maria A. Mikhalkova — Junior Researcher.

Saint Petersburg, 199034, sc 57218284288

mikhalkova_maria@infran.ruhttps://orcid.org/0000-0002-7239-2096ЯчнаяВ. О.YachnayaV. O.

Ячная Валерия Олеговна — младший научный сотрудник, Институт физиологии имени И.П. Павлова РАН; аспирант, Санкт-Петербургский ГУ аэрокосмического приборостроения

Санкт-Петербург, 199034, 190000, sc 57209076316

Valeria O. Yachnaya — Junior Researcher, Pavlov Institute of Physiology of the Russian Academy of Sciences; PhD Student, Saint Petersburg SU of Aerospace Instrumentation.

Saint Petersburg, 199034, 190000, sc 57209076316

yachnaya_valeria@infran.ruСанкт-Петербургский государственный университет аэрокосмического приборостроенияРоссияSaint Petersburg State University of Aerospace InstrumentationRussian FederationИнститут физиологии имени И.П. Павлова РАНРоссияPavlov Institute of Physiology of the Russian Academy of SciencesRussian FederationСанкт-Петербургский государственный университет аэрокосмического приборостроения; Институт физиологии имени И.П. Павлова РАНРоссияSaint Petersburg State University of Aerospace Instrumentation; Pavlov Institute of Physiology of the Russian Academy of SciencesRussian Federation202524042025252273285Copyright © Луцив В.Р., Михалькова М.А., Ячная В.О., 20252025Луцив В.Р., Михалькова М.А., Ячная В.О.Lutsiv V.R., Mikhalkova M.A., Yachnaya V.O.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://ntv.elpub.ru/jour/article/view/447Введение

Введение. Наблюдаемый мир иерархически структурирован. Механизм взаимодействия встречных информационных потоков в иерархически организованной системе был назван «адаптивным резонансом» и успешно промоделирован полвека назад в искусственной нейронной сети, анализирующей обонятельные стимулы, а затем — при распознавании изображений. В дальнейшем полезный потенциал совместного применения в нейронных сетях принципов структурного анализа и адаптивного резонанса долго не использовался. В последние годы эти принципы были применены в капсульных сетях и позволили превзойти результаты, достигнутые другими моделями нейронных сетей. Это показало необходимость систематизировать использованные пути их реализации.

Метод

Метод. Выполнен анализ научно-технических работ, опубликованных за последние 50 лет по применению принципов структурного анализа и адаптивного резонанса при обработке изображений в искусственных нейронных сетях. Основные результаты. Сравнительный анализ подтвердил эффективность применения этих принципов при решении задач автоматической обработки изображений и выявил пути их наиболее эффективной реализации в искусственных нейронных сетях.

Обсуждение

Обсуждение. В связи с успехами, достигнутыми сверточными сетями при распознавании изображений, их разработчики не применяли связанные с особенностями организации наблюдаемого мира принципы структурного анализа и адаптивного резонанса. Однако использование данных принципов позволяет более эффективно решать задачи обработки изображений, и дальнейшие исследования в области искусственных нейронных сетей целесообразно проводить в этом направлении.

The observable world is hierarchically structured. The interaction technique of counter motion of information flows in the hierarchically organized systems was named “the adaptive resonance” and successfully modeled in artificial neural network of olfactory stimuli analysis and then applied for image recognition. The usefulness of application of the principles of hierarchical structural analysis and adaptive resonance was then forgotten for a long time. Recently, these principles were applied again in the capsule neural networks that outperformed the best modern models of other neural networks. This shows the necessity of systematizing the ways of practical implementation of these principles. The experience of application of structural analysis and adaptive resonance in the tasks of image recognition by artificial neural networks was inspected through the scientific and technical materials published last half-century. The comparative analysis carried out confirmed the efficiency of application of these principles in automatic image processing. The methods of most efficient realization of structural analysis and adaptive resonance in artificial neural networks were also determined. While the successful results of image recognition were reached by convolutional neural networks, their developers consigned to oblivion the principles of structural analysis and adaptive resonance following from the organization peculiarities of observable environment. However, the come-back to application of these principles would result in additional success in solving the tasks of image processing; thus, the further investigation of artificial neural networks is worth to be carried out in this area.

искусственные нейронные сетиструктурный анализадаптивный резонансиерархическая организациясверточные сетикапсульные сетитрансформерыartificial neural networksstructural analysisadaptive resonancehierarchical organizationconvolutional networkscapsule networkstransformersReferencesGrossberg S. Adaptive pattern classification and universal recoding: II. Feedback, expectation, olfaction, illusions // Biological Cybernetics. 1976. V. 23. N 4. P. 187–202. https://doi.org/10.1007/BF00340335Grossberg S. Adaptive pattern classification and universal recoding: II. Feedback, expectation, olfaction, illusions. Biological Cybernetics, 1976, vol. 23, no. 4, pp. 187–202. https://doi.org/10.1007/BF00340335Carpenter G.A., Grossberg S. Category learning and adaptive pattern recognition: a neural network model // Proc. of the 3rd Army Conference on Applied Mathematics and Computing. 1986. P. 37–56.Carpenter G.A., Grossberg S. Category learning and adaptive pattern recognition: a neural network mode. Proc. of the 3rd Army Conference on Applied Mathematics and Computing, 1986, pp. 37–56.Carpenter G.A., Grossberg S. ART 2: Self-organization of stable category recognition codes for analog input patterns // Applied Optics. 1987. V. 26. N 12. P. 4919–4930. https://doi.org/10.1364/AO.26.004919Carpenter G.A., Grossberg S. ART 2: Self-organization of stable category recognition codes for analog input patterns. Applied Optics, 1987, vol. 26, no. 12, pp. 4919–4930. https://doi.org/10.1364/AO.26.004919Rumelhart D.E. Learning Internal Representations by Error Propagation. Institute for Cognitive Science, 1985. 34 p.Rumelhart D.E. Learning Internal Representations by Error Propagation. Institute for Cognitive Science, 1985, 34 p.Yamada K., Kami H., Tsukumo J., Temma T. Handwritten numeral recognition by multilayered neural network with improved learning algorithm // Proc. of the International 1989 Joint Conference on Neural Networks. 1989. P. 259–266. https://doi.org/10.1109/IJCNN.1989.118708Yamada K., Kami H., Tsukumo J., Temma T. Handwritten numeral recognition by multilayered neural network with improved learning algorithm. Proc. of the International 1989 Joint Conference on Neural Networks , 1989, pp. 259–266. https://doi.org/10.1109/IJCNN.1989.118708Fukushima K. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position // Biological Cybernetics. 1980. V. 36. N 4. P. 193–202. https://doi.org/10.1007/bf00344251Fukushima K. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biological Cybernetics, 1980, vol. 36, no. 4, pp. 193–202. https://doi.org/10.1007/bf00344251Krizhevsky A., Sutskever I., Hinton G.E. ImageNet classification with deep convolutional neural networks // Proc. of the 26th International Conference on Neural Information Processing Systems (NIPS'12). 2012. V. 1. P. 1097–1105.Krizhevsky A., Sutskever I., Hinton G.E. ImageNet classification with deep convolutional neural networks. Proc. of the 26th International Conference on Neural Information Processing Systems (NIPS'12), 2012, vol. 1, pp. 1097–1105.Fukushima K. A neural network model for selective attention in visual pattern recognition // Biological Cybernetics. 1986. V. 55. N 1. P. 5–15. https://doi.org/10.1007/BF00363973Fukushima K. A neural network model for selective attention in visual pattern recognition. Biological Cybernetics, 1986, vol. 55, no. 1, pp. 5–15. https://doi.org/10.1007/BF00363973Hopfield J.J., Tank D.W. Neural computation of decisions in optimization problems // Biological Cybernetics. 1985. V. 52. N 3. P. 141–152. https://doi.org/10.1007/bf00339943Hopfield J.J., Tank D.W. Neural computation of decisions in optimization problems. Biological Cybernetics, 1985, vol. 52, no. 3, pp. 141–152. https://doi.org/10.1007/bf00339943Lutsiv V.R., Novikova T.A. On the use of neurocomputer for stereoimage processing // Pattern Recognition and Image Analysis // Advances in Mathematical Theory and Applications. 1992. V. 2. N 4. P. 441–444.Lutsiv V.R., Novikova T.A. On the use of neurocomputer for stereoimage processing. Pattern Recognition and Image Analysis. Advances in Mathematical Theory and Applications, 1992, vol. 2, no. 4, pp. 441–444.Данилов Е.П., Луцив В.Р., Новикова Т.А., Малышев И.А. Разработка методов и алгоритмов реализации интеллектуальных вычислительных устройств с использованием нейронных сетей в интересах задач анализа и сопоставления изображений. Отчет о работе 22201-007-94, Шифр «Нейросеть». СПб.: Государственный оптический институт им. С.И. Вавилова, С.-Петербург, 1994. С. 39–50.Danilov E.P., Lutciv V.R., Novikova T.A., Malyshev I.A. Development of methods and algorithms for the implementation of intelligent computing devices using neural networks for the purposes of image analysis and comparison problems. Statement of work no. 22201-007-94, Code “Neural network”. St. Petersburg, Open Joint Stock Company «S. I. Vavilov State Optical Institute», 1994, pp. 39–50. (in Russian)Dolinov D.S., Zherebko A.K., Lutsiv V.R., Novikova T.A. Using artificial neural networks in image processing problems //Journal of Optical Technology. 1997. V. 64. N 2. P. 112–118.Dolinov D.S., Zherebko A.K., Lutsiv V.R., Novikova T.A. Using artificial neural networks in image processing problems. Journal of Optical Technology, 1997, vol. 64, no. 2, pp. 112–118.Lutsiv V.R., Malyshev I.A., Pepelka V.A. Automatic fusion of multiple-sensor and multiple-season images // Proceedings of SPIE. 2001. V. 4380. P. 174–183. https://doi.org/10.1117/12.436990Lutsiv V.R., Malyshev I.A., Pepelka V.A. Automatic fusion of multiple-sensor and multiple-season images. Proceedings of SPIE, 2001, vol. 4380, pp. 174–183. https://doi.org/10.1117/12.436990Lutsiv V.R., Malyshev I.A., Pepelka V.A., Potapov A.S. The Target independent algorithms for description and structural matching of aerospace photographs // Proceedings of SPIE. 2002. V. 4741. P. 351–362. https://doi.org/10.1117/12.478732Lutsiv V.R., Malyshev I.A., Pepelka V.A., Potapov A.S. The Target independent algorithms for description and structural matching of aerospace photographs. Proceedings of SPIE, 2002, vol. 4741, pp. 351–362. https://doi.org/10.1117/12.478732Луцив В. Автоматический анализ изображений: Объектно-независимый структурный подход. Саарбрюккен: Ламберт Академик Паблишинг, 2011. 308 с.Lutsiv V. Automatic image analysis: Object-independent structured approach. Lambert Academic Publishing, 2011, 308 p. (in Russian)Слэйгл Д.P. Искусственный интеллект: подход на основе эвристического программирования. М.: Мир, 1973. 319 с.Slagle J.R. Artificial Intelligence: The Heuristic Programming Approach. McGraw Hill, 1973, 196 p.Lutsiv V.R., Malyshev I.A., Potapov A.S. Hierarchical structural matching algorithms for registration of aerospace images // Proceedings of SPIE. 2003. V. 5238. P. 164–175. https://doi.org/10.1117/12.511770Lutsiv V.R., Malyshev I.A., Potapov A.S. Hierarchical structural matching algorithms for registration of aerospace images. Proceedings of SPIE, 2003, vol. 5238, pp. 164–175. https://doi.org/10.1117/12.511770Ponomarev S., Lutsiv V., Malyshev I. Automatic structural matching of 3D image data // Proceedings of SPIE. 2015. V. 9649. P. 96490M. https://doi.org/10.1117/12.2194312Ponomarev S., Lutsiv V., Malyshev I. Automatic structural matching of 3D image data. Proceedings of SPIE, 2015, vol. 9649, pp. 96490M. https://doi.org/10.1117/12.2194312Girshick R., Iandola F., Darrell T., Malik J. Deformable part models are convolutional neural networks // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2015. P. 437–446. https://doi.org/10.1109/cvpr.2015.7298641Girshick R., Iandola F., Darrell T., Malik, J. Deformable part models are convolutional neural networks. Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 437–446. https://doi.org/10.1109/cvpr.2015.7298641Felzenszwalb P.F., Girshick R.B., McAllester D., Ramanan D. Object detection with discriminatively trained part-based models // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2010. V. 32. N 9. P. 1627–1645. https://doi.org/10.1109/tpami.2009.167Felzenszwalb P.F., Girshick R.B., McAllester D., Ramanan D. Object detection with discriminatively trained part-based models. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, vol. 32, no. 9, pp. 1627–1645. https://doi.org/10.1109/tpami.2009.167Yang Y., Ramanan D. Articulated human detection with flexible mixtures of parts // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2013. V. 35. N 12. P. 2878–2890. https://doi.org/10.1109/tpami.2012.261Yang Y., Ramanan D. Articulated human detection with flexible mixtures of parts. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, vol. 35, no. 12, pp. 2878–2890. https://doi.org/10.1109/tpami.2012.261Wang Y., Tran D., Liao Z. Learning hierarchical poselets for human parsing // CVPR 2011. P. 1705–1712. https://doi.org/10.1109/cvpr.2011.5995519Wang Y., Tran D., Liao Z. Learning hierarchical poselets for human parsing. CVPR 2011, pp. 1705–1712. https://doi.org/10.1109/cvpr.2011.5995519Wan L., Eigen D., Fergus R. End-to-end integration of a convolutional network, deformable parts model and non-maximum suppression // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2015. P. 851–859. https://doi.org/10.1109/cvpr.2015.7298686Wan L., Eigen D., Fergus R. End-to-end integration of a convolutional network, deformable parts model and non-maximum suppression. Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 851–859. https://doi.org/10.1109/cvpr.2015.7298686Girshick R., Donahue J., Darrell T., Malik J. Rich feature hierarchies for accurate object detection and semantic segmentation // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition. 2014. P. 580–587. https://doi.org/10.1109/CVPR.2014.81Girshick R., Donahue J., Darrell T., Malik J. Rich feature hierarchies for accurate object detection and semantic segmentation. Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2014, pp. 580–587. https://doi.org/10.1109/CVPR.2014.81Hinton G.E., Krizhevsky A., Wang S.D. Transforming autoencoders // Lecture Notes in Computer Science. 2011. V. 6791. P. 44–51. https://doi.org/10.1007/978-3-642-21735-7_6Hinton G.E., Krizhevsky A., Wang S.D. Transforming auto-encoders. Lecture Notes in Computer Science, 2011, vol. 6791, pp. 44–51. https://doi.org/10.1007/978-3-642-21735-7_6Sabour S., Frosst N., Hinton G.E. Dynamic routing between capsules // arXiv. 2017. arXiv:1710.09829v2. https://doi.org/10.48550/arXiv.1710.09829Sabour S., Frosst N., Hinton G.E. Dynamic routing between capsules. arXiv, 2017, arXiv:1710.09829v2. https://doi.org/10.48550/arXiv.1710.09829Hinton G., Sabour S., Frosst N. Matrix capsules with EM routing // Proc. of the 6th International Conference on Learning Representations (ICLR 2018). 2018. P. 1–15.Hinton G., Sabour S., Frosst N. Matrix capsules with EM routing. Proc. of the 6th International Conference on Learning Representations (ICLR 2018), 2018, pp. 1–15.LeCun Y., Huang F.J., Bottou L. Learning methods for generic object recognition with invariance to pose and lighting // Proc. of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2004). 2004. P. 97–104. https://doi.org/10.1109/cvpr.2004.1315150LeCun Y., Huang F.J., Bottou L. Learning methods for generic object recognition with invariance to pose and lighting. Proc. of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2004), 2004, pp. 97–104. https://doi.org/10.1109/cvpr.2004.1315150Kosiorek A.R., Sabour S., Teh Y.W., Hinton G.E. Stacked capsule autoencoders // arXiv. 2019. arXiv:1906.06818v2. https://doi.org/10.48550/arXiv.1906.06818Kosiorek A.R., Sabour S., Teh Y.W., Hinton G.E. Stacked capsule autoencoders. arXiv, 2019, arXiv:1906.06818v2. https://doi.org/10.48550/arXiv.1906.06818Lee J., Lee Y., Kim J., Kosiorek A.R., Choi S., Teh Y.W. Set Transformer: a framework for attention-based permutation-invariant neural networks // arXiv. 2019. arXiv:1810.00825v3. https://doi.org/10.48550/arXiv.1810.00825Lee J., Lee Y., Kim J., Kosiorek A.R., Choi S., Teh Y.W. Set Transformer: a framework for attention-based permutation-invariant neural networks. arXiv, 2019, arXiv:1810.00825v3. https://doi.org/10.48550/arXiv.1810.00825Haeusser P., Plapp J., Golkov V., Aljalbout E., Cremers D. Associative deep clustering: training a classification network with no labels // Lecture Notes in Computer Science. 2019. V. 11269. P. 18–32. https://doi.org/10.1007/978-3-030-12939-2_2Haeusser P., Plapp J., Golkov V., Aljalbout E., Cremers D. Associative deep clustering: training a classification network with no labels. Lecture Notes in Computer Science, 2019, vol. 11269, pp. 18–32. https://doi.org/10.1007/978-3-030-12939-2_2Sun W., Tagliasacchi A., Deng B., Sabour S., Yazdani S., Hinton G., Yi K.M. Canonical capsules: self-supervised capsules in canonical pose // arXiv. 2021. arXiv:2012.04718v2. https://doi.org/10.48550/arXiv.2012.04718Sun W., Tagliasacchi A., Deng B., Sabour S., Yazdani S., Hinton G., Yi K.M. Canonical capsules: self-supervised capsules in canonical pose. arXiv, 2021, arXiv:2012.04718v2. https://doi.org/10.48550/arXiv.2012.04718Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A.N., Kaiser L., Polosukhin I. Attention is all you need // arXiv. 2017. arXiv:1706.03762v5. https://doi.org/10.48550/arXiv.1706.03762Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A.N., Kaiser L., Polosukhin I. Attention is all you need. arXiv, 2017, arXiv:1706.03762v5. https://doi.org/10.48550/arXiv.1706.03762Parmar N., Vaswani A., Uszkoreit J., Kaiser L., Shazeer N., Ku A., Tran D. Image Transformer // arXiv. 2018. arXiv:1802.05751v3. https://doi.org/10.48550/arXiv.1802.05751Parmar N., Vaswani A., Uszkoreit J., Kaiser L., Shazeer N., Ku A., Tran D. Image Transformer. arXiv, 2018, arXiv:1802.05751v3. https://doi.org/10.48550/arXiv.1802.05751Dosovitskiy A., Beyer L., Kolesnikov A., Weissenborn D., Zhai X., Unterthiner T., Dehghani M., Minderer M., Heigold G., Gelly S., Uszkoreit J., Houlsby N. An image is worth 16x16 words: transformers for image recognition at scale // arXiv. 2021. https://doi.org/10.48550/arXiv.2010.11929Dosovitskiy A., Beyer L., Kolesnikov A., Weissenborn D., Zhai X., Unterthiner T., Dehghani M., Minderer M., Heigold G., Gelly S., Uszkoreit J., Houlsby N. An image is worth 16x16 words: transformers for image recognition at scale. arXiv, 2021, https://doi.org/10.48550/arXiv.2010.11929Hinton G. How to represent part-whole hierarchies in a neural network // Neural Computation. 2023. V. 35. N 3. P. 413–452. https://doi.org/10.1162/neco_a_01557Hinton G. How to represent part-whole hierarchies in a neural network. Neural Computation, 2023, vol. 35, no. 3, pp. 413–452. https://doi.org/10.1162/neco_a_01557Kohonen T. Self-organization and Associative Memories. Springer, 1984. 255 p.Kohonen T. Self-Organization and Associative Memories. Springer, 1984, 255 p.Луцив В.Р., Михалькова М.А., Ячная В.О. Компьютерное зрение: учеб. пособие: в 3 ч. Ч. 3. Современные модификации архитектур нейронных сетей и методы улучшения изображений. СПб.: ГУАП, 2024. 196 с.Lutciv V.R., Mikhalkova M.A., Iachnaia V.O. Computer Vision. A Tutorial in 3 parts. Part 3: Modern Modifications of Neural Network Architectures and Methods of Image Enhancement. St. Petersburg, GUAP, 2024, 196 p. (in Russian)

The authors declare that there are no conflicts of interest present.