COMPARING TYPES OF KERNELS IN CONVOLUTIONAL LAYERS OF NEURAL NETWORKS
V. M. Giniyatullin,
A. V. Khlybov,
M. A. Fedorov,
T. A. Asadullin,
A. S. Krutin,
I. A. Osipov,
D. M. Zaripov https://doi.org/10.34822/1999-7604-2022-3-84-98
Abstract
The article studies the process of using fixed specified convolutional kernels. The calculations are conducted considering the change in filter size and stride. In the course of the study, four different sets of images and structures of neural networks are selected. The article discusses four types of specified kernels: vertical, horizontal, diagonal, and ring. The dependency of the accuracy of image recognition in monochrome and color is analyzed. A positive shift is observed when training fixed kernels.
About the Authors
V. M. Giniyatullin
Ufa State Oil Technical University, Ufa
Russian Federation
Russian Federation
Candidate of Sciences (Engi-neering), Associate Professor
E-mail: fentazer@mail.ru
A. V. Khlybov
Ufa State Oil Technical University, Ufa
Russian Federation
Russian Federation
Postgraduate
E-mail: brinkinvision@gmail.com
M. A. Fedorov
Ufa State Oil Technical University, Ufa
Russian Federation
Russian Federation
Student
E-mail: MasterOfHoMM@gmail.com
T. A. Asadullin
Ufa State Oil Technical University, Ufa
Russian Federation
Russian Federation
Student
E-mail: tealredplanet@gmail.com
A. S. Krutin
Ufa State Oil Technical University, Ufa
Russian Federation
Russian Federation
Student
E-mail: krut_inuly@mail.ru
I. A. Osipov
Ufa State Oil Technical University, Ufa
Russian Federation
Russian Federation
Student
E-mail: warluswarlusgg@gmail.com
D. M. Zaripov
Ufa State Oil Technical University, Ufa
Russian Federation
Russian Federation
Candidate of Sciences (Physics and Mathematics), Associate Professor
E-mail: damir.zaripov@gmail.com
References
1. Mahajan P. Fully Connected vs Convolutional Neural Networks. URL: https://medium.com/swlh/fully-connected-vs-convolutional-neural-networks-813ca7 bc6ee5 (дата обращения: 25.07.2022).
2. Lin Z., Memisevic R., Konda K. How Far Can We Go without Convolution: Improving Fully-Connected Networks // arXiv. 2015. URL: https://arxiv.org/abs/ 1511.02580 (дата обращения: 25.07.2022).
3. Vajpayee S. License Plate Recognition Using CNN. URL: https://www.kaggle.com/code/sarthakvajpayee/ license-plate-recognition-using-cnn/notebook (дата обращения: 25.07.2022).
4. Гиниятуллин В. М., Ермолаев Е. В., Хлыбов А. В. Формализация процедуры подбора сверточных ядер // Вест. кибернетики. 2022. № 2. С. 66–74.
5. Хлыбов А. В., Ермолаев Е. В., Зеленов А. С., Коняева Е. Э., Фаррахова К. А., Хужиахметов В. Э., Яковлев А. С. Преобразование весов матриц сверточных нейронных сетей // Теория и практика процессов хим. технологии (Марушкинские чтения) : материалы VI Междунар. науч. конф. Уфа, 2021. С. 269–270.
6. CNN Fruit Classification. Data. URL: https://www.kag gle.com/etatbak/cnn-fruit-classification/ data (дата обращения: 25.07.2022).
7. Pneumonia Detection Using CNN (92.6 % Accura-cy). Data. URL: https://www.kaggle.com/madz2000/ pneumonia-detection-using-cnn-92-6-accuracy/data (дата обращения: 25.07.2022).
8. Detecting Malaria | Keras | CNN. Data. URL: https://www.kaggle.com/miracle9to9/detecting-malaria-keras-cnn/data (дата обращения: 25.07.2022).
9. Fashion MNIST – Keras CNN. Data. URL: https://www.kaggle.com/vishnu0399/fashion-mnist-keras-cnn/data (дата обращения: 25.07.2022).
10. CNN Fruit Classification. Notebook. URL: https://www.kaggle.com/etatbak/cnn-fruit-classification/ notebook (дата обращения: 25.07.2022).
11. Pneumonia Detection Using CNN (92.6 % Accuracy). Notebook. URL: https://www.kaggle.com/ madz2000/pneumonia-detection-using-cnn-92-6-accuracy/notebook (дата обращения: 25.07.2022).
12. Detecting Malaria | Keras | CNN. Notebook. URL: https://www.kaggle.com/miracle9to9/detecting-malaria-keras-cnn/notebook (дата обращения: 25.07.2022).
13. Fashion MNIST – Keras CNN. Notebook. URL: https://www.kaggle.com/vishnu0399/fashion-mnist-keras-cnn/notebook (дата обращения: 25.07.2022).
14. Understand How Color to Gray Scale Works Using OpenCV. URL: https://www.learnpythonwithrune.org/ understand-how-color-to-gray-scale-works-using-opencv/ (дата обращения: 25.07.2022).
15. Tf.keras.layers.Dropout. URL: https://www.tensor flow.org/api_docs/python/tf/keras/layers/Dropout (дата обращения: 25.07.2022).
Review
For citations:
Giniyatullin V.M., Khlybov A.V., Fedorov M.A., Asadullin T.A., Krutin A.S., Osipov I.A., Zaripov D.M. COMPARING TYPES OF KERNELS IN CONVOLUTIONAL LAYERS OF NEURAL NETWORKS. Proceedings in Cybernetics. 2022;(3 (47)):84-98. (In Russ.) https://doi.org/10.34822/1999-7604-2022-3-84-98
Views:
157
Email this article 