Yaloveha V. Multispectral images processing methods in a computerized system based on deep learning neural networks

Українська версія

Thesis for the degree of Doctor of Philosophy (PhD)

State registration number

0824U001465

Thesis Registration Form

0824U001465.pdf

Applicant for

Specialization

  • 123 - Комп’ютерна інженерія

24-05-2024

Specialized Academic Board

ДФ 64.050.133-5263

National Technical University "Kharkiv Polytechnic Institute"

Essay

The dissertation work is devoted to the solution of an actual scientific and technical problem of multispectral satellite images processing based on deep learning neural networks in a computerized system to increase the performance and solutions quality that will have the possibility of obtaining actual information about Earth's surface objects during Earth remote sensing. The purpose of the dissertation is to increase the classification quality of actual information about Earth's surface objects during remote sensing by developing new and improving existing multispectral satellite images classification methods based on deep learning methods. The object of research is the process of Earth remote sensing multispectral satellite images multiclass classification in a computer system. The subject of research is methods and means of multiclass classification in a computer system based on deep learning methods. The introduction substantiates the relevance of a scientific and technical problem of Earth's surface multispectral images processing in a computerized system, presents the connection of the work with scientific programs, plans, and topics, provides scientific novelty, presents the practical significance of the obtained results and provides the applicant personal contribution information with publications on the topic of the dissertation. The scientific novelty of the results. As a result of the dissertation work, the following scientific results were obtained within this area: 1. For the first time, a method of Earth's surface multispectral satellite multiclass classification images is proposed, which differs from the known procedure of finding the optimal set of spectral indexes based on the proposed architecture of a convolutional neural network in a computerized system, which increased the Earth's surface objects classification accuracy. 2. The method of multispectral multiclass images classification based on a convolutional neural network with spectral indexes has been improved by optimizing the proposed procedure of rough-tuning and fine-tuning stages under given budget restrictions, which on the one hand increased the satellite images classification quality result metrics of a convolutional neural network, and on the other hand, considered existing resource limitations. 3. The method of Earth's cover high-resolution multispectral satellite multiclass classification images was improved, which differs from the known by transfer learning of convolutional neural networks based on the proposed high-resolution EuroPlanet dataset and the optimal spectral indexes configuration and made increased the Earth remote sensing data classification accuracy and the neural network model performance on the Ukrainian territory. The developed and improved methods are the scientific and methodological basis for the designing algorithms and software. The practical results include the following: – the method and software for designing a convolutional neural network were developed for the task of multispectral multiclass satellite images classification with an optimal set of spectral indexes, which increased the classification accuracy up to 84.19% and the F1 metric up to 84.05%; – the improved convolutional neural networks optimization method and software have been developed for the task of the Earth’s surface satellite images, which increased the classification accuracy and F1 metric to 97.04% and 97.05% respectively, and for the classes Herbaceous Vegetation, Permanent Crop, and Highway the F1 metric on the test dataset increased up to 20%. In addition, the use of the modern Ray Tune framework made it possible to effectively use the available resources under the defined budget restrictions; – the high-quality high-resolution satellite images filtering procedure has been developed, which accelerated and automated the creation of the EuroPlanet dataset in a computerized system; – the multiclass land cover EuroPlanet high-resolution images with the optimal configuration of spectral indexes classification method has been improved and software has been developed. The classification accuracy on the test data increased to 93.83%, and the F1 metric increased to 93.56%. The practical possibility of using the improved method is shown. The research results confirmed the practical and theoretical significance of the developed methods and procedures, provided practical recommendations for the application of the developed methods in the conditions of a full-scale Russian invasion of Ukraine, and considered the prospects for their further development.

Read more

Research papers

1. V. Yaloveha, A. Podorozhniak, H. Kuchuk, and N. Garashchuk, "Performance comparison of CNNs on high-resolution multispectral dataset applied to land cover classification problem", Radioelectronic and Computer Systems, 2023, no. 2, pp. 107-118. (А)

2. V. Yaloveha, A. Podorozhniak, and H. Kuchuk, "Convolutional neural network hyperparameter optimization applied to land cover classification", Radioelectronic and Computer Systems, 2022, no. 1, pp. 115-128. (А)

3. V. Yaloveha, D. Hlavcheva, and A. Podorozhniak, "Spectral Indexes Evaluation for Satellite Images Classification using CNN", Journal of Information and Organizational Sciences, 2021, vol. 46, no. 2, pp. 95-113. (Scopus, Хорватія)

4. H. Kuchuk, D. Hlavcheva, A. Podorozhnіak, and V. Yaloveha, "Application of Deep Learning in the Processing of the Aerospace System’s Multispectral Images", Handbook of Research on Artificial Intelligence Applications in the Aviation and Aerospace Industries, IGI Global, 2020, pp. 134-147. (Монографія. Розділ 5.)

5. D. Hlavcheva, V. Yaloveha, A. Podorozhniak, and N. Lukova-Chuiko, "A comparison of classifiers applied to the problem of biopsy images analysis", Advanced Information Systems, Kharkiv, 2020, vol. 4, no. 2, pp. 12-16. (Б)

6. V. Yaloveha, D. Hlavcheva, and A. Podorozhniak, "Usage of convolutional neural network for multispectral image processing applied to the problem of detecting fire hazardous forest areas", Advanced Information Systems, Kharkiv, 2019, vol. 3, no. 1, pp. 116-120. (Б)

7. D. Hlavcheva, V. Yaloveha, and A. Podorozhniak, "Application of convolutional neural network for histopathological analysis", Advanced Information Systems, Kharkiv, 2019, vol. 3, no. 4, pp. 69-73. (Б)

8. Д. М. Главчева та В. А. Яловега, "Капсульні нейронні мережі", Системи управління, навігації та зв’язку, Полтава, 2018, Т. 5, № 51, с. 132-135. (Б)

9. V. Yaloveha, A. Podorozhniak, H. Kuchuk, T. Orlova, V. Noskov, and V. Gorbulik, "Modern Applications of High-Resolution Multispectral EuroPlanet Dataset", 2023 IEEE 4th KhPI Week on Advanced Technology (KhPIWeek), Kharkiv, Ukraine, 2023, pp. 166-170. (Scopus)

10. В. А. Яловега та А. О. Подорожняк, "Сучасні методи отримання мультиспектральних зображень високої роздільної здатності", Матеріали XХХІ Міжнар. науково-практ. конф. (MicroCAD-2023) Інформаційні технології: наука, техніка, технологія, освіта, здоров’я, Харків, 2023, c. 1218.

11. V. Yaloveha, А. Podorozhniak, "Transfer Learning Technique Applied to Multispectral Images Classification Problem", Матеріали XIII Міжнар. науково-техн. конф. Сучасні напрями розвитку інформаційно-комунікаційних технологій та засобів управління, Баку-Харків-Жиліна, 2023, с. 27.

12. А. О. Подорожняк та В. А. Яловега, "Сучасні алгоритми оптимізації згорткових штучних нейронних мереж", Матеріали IX Міжнар. науково-техн. конф. (ІУШІ-2022) Інформатика, управління та штучний інтелект, Харків, 2022, с. 109.

13. D. Hlavcheva, V. Yaloveha, A. Podorozhniak, and H. Kuchuk, "Comparison of CNNs for Lung Biopsy Images Classification", 2021 IEEE 2nd Ukraine Conference on Electrical and Computer Engineering (UKRCON), Lviv, Ukraine, 2021, pp. 1-5. (Scopus)

14. V. Yaloveha, D. Hlavcheva, and A. Podorozhniak, "Modern hyperparameter optimization approaches in Deep Learning", Матеріали XI Міжнар. науково-техн. конф. Сучасні напрями розвитку інформаційно-комунікаційних технологій та засобів управління, Харків, 2021, с. 13.

15. V. Yaloveha, D. Hlavcheva, and A. Podorozhniak, "CNN Hyperparameters Optimization Applied to EuroSAT Dataset", Матеріали VIII Міжнар. науково-техн. конф. (ІУШІ-2021) Інформатика, управління та штучний інтелект, Харків, 2021, с. 113.

16. D. Hlavcheva, V. Yaloveha, and A. Podorozhniak, "Using of Deep Learning Neural Networks for Biopsy Images Classification", Матеріали ХV Міжнар. науково-практ. конф. магістрантів та аспірантів «Теоретичні та практичні дослідження молодих науковців», Харків, 2021, с. 12.

17. D. Hlavcheva, V. Yaloveha, A. Podorozhniak, and H. Kuchuk, "Tumor Nuclei Detection in Histopathology Images Using R-CNN", Proceedings of the 16th International Conference on ICT in Education, Research and Industrial Applications (ICTERI 2020), Kharkiv, Ukraine, 2020, vol. 2740, pp. 63-74. (Scopus)

18. V. Yaloveha, D. Hlavcheva, and A. Podorozhniak, "Modern high-resolution satellite image processing overview", Матеріали Матеріали VІ Міжнар. науково-техн. конф. Проблеми інформатизації, Черкаси, 2020, с. 20.

19. V. Yaloveha, D. Hlavcheva, A. Podorozhniak, and H. Kuchuk, "Fire hazard research of forest areas based on the use of convolutional and capsule neural networks", 2019 IEEE 2nd Ukraine Conference on Electrical and Computer Engineering (UKRCON), Lviv, Ukraine, 2019, pp. 828-832. (Scopus)

20. Я. Р. Широкорад, Д. М. Главчева та В. А. Яловега, "Перенавчання нейронних мереж", Матеріали ІХ Міжнар. науково-техн. конф. Сучасні напрями розвитку інформаційно-комунікаційних технологій та засобів управління, Харків, 2019, с. 75.

21. Д. М. Главчева, В. А. Яловега та Я. Р. Широкорад, "Створення індексних зображень", Матеріали ІХ Міжнар. науково-техн. конф. Сучасні напрями розвитку інформаційно-комунікаційних технологій та засобів управління, Харків, с. 28, 2019.

22. Д. М. Главчева, В. А. Яловега та А. О. Подорожняк, "Дослідження пожежонебезпечності лісових територій на основі використання капсульних та згорткових нейронних мереж", Матеріали Всеукраїнської науково-практ. конф. молодих науковців і студентів «Інтелектуальний потенціал – 2019», Хмельницький, 2019, с. 14-17.

23. D. M. Hlavcheva, and V. A. Yaloveha, "CapsNet versus ConvNet", Матеріали V Міжнар. науково-техн. конф. (ІУШІ-2018) Інформатика, управління та штучний інтелект, Харків, 2018, с. 22-23.

24. В. А. Яловега, Д. М. Главчева, А. О. Подорожняк та Г. А. Кучук, "Комп’ютерна програма для обробки мультиспектральних зображень та навчання згорткової та капсульної нейронних мереж", Авторське право на твір № 87363, 2019, заявл. 01.04.2019, опубл. 26.07.2019, Бюл. № 53.

Read more

Files

autoreferat-tytul_dysertatsiia_2024_Yaloveha_Multispectral_images_processing.pdf

dysertatsiia_2024_Yaloveha_Multispectral_images_processing.pdf

Similar theses

0824U000994

Bohdan Khudik

Methodology for building a recommendation system for the field of education based on a hybrid approach

0824U000993

Volodymyr V. Dmytrenko

Model and method of building convergent software-configured networks with guaranteed reliability of control information.

0824U000864

Mykola M. Gertsiuk

Information technology for river pollution forecasting in emergencies based on regression methods and heuristic algorithms

0824U000347

Hanna S. Yelisieieva

Recurrent logarithmic analog-to-digital converters

0824U000366

Dmytro O. Kushnir

Methods and means of searching and recognizing objects in video images on the mobile platform in real-time.