N.A. Andriyanov1, A.L. Kim2

1,2 Financial University under the Government of the Russian Federation (Moscow, Russia)

1 naandriyanov@fa.ru, 2 alkim@fa.ru

Currently, it is quite a difficult task to improve the quality of neural network detectors. This is primarily due to the impossibility of repeatedly increasing the training samples. At the same time, the well-known augmentation methods, although providing some increase in accuracy, have saturation limits. In the tasks of satellite images analysis for fire monitoring, each share of the accuracy percentage gained plays a special role. In this connection it is necessary to consider non-standard augmentation methods.

The main research objective of this paper is to improve the accuracy of fire detection by augmenting the training sample using generative image models.

In this study, solutions based on open cloud-based image generation services are considered. Different object detector architectures are proposed and compared in terms of accuracy. The results show that using generative models to expand the training sample is on average 1–2% more efficient than using standard data augmentation methods. Moreover, the paper presents a combined approach for training image base augmentation. The combination of generative models and traditional augmentations can improve object detection accuracy by 4–5% compared to using standard augmentations alone. Thus, the combined method of training sample expansion using both generative models and traditional augmentations demonstrates the best results in the task of improving the accuracy of fire detection on satellite images. This approach can be effectively applied in practical computer vision systems to improve the quality of object detection.

The obtained results of the study will be useful for specialists in the field of computer vision, and the developed models can be applied by special services when monitoring forest territory using satellite images.

Andriyanov N.A., Kim A.L. Efficient fire detection using generative augmentations. Neurocomputers. 2024. V. 26. № 6. Р. 14-22. DOI: https://doi.org/10.18127/j19998554-202406-03 (In Russian)

1. Yuzugullu O., Lorenz F., Fröhlich P., Liebisch F. Understanding Fields by Remote Sensing: Soil Zoning and Property Mapping. Remote Sensing. 2020. V. 12. № 7. P. 1116. DOI 10.3390/rs12071116.
2. Firsov N.M., Khabibullin A.F., Sashchenko V.N., Malitsky R.B., Platonov E.Yu. Experience in detection and monitoring of forest fires using space images. Forests of Russia and management. 2019. № 4(71). P. 33–41. (in Russian)
3. Andriyanov N.A., Dement'ev V.E. Application of mixed models of random fields for the segmentation of satellite images. CEUR Workshop Proceedings. 2018. V. 2210. P. 219–226. DOI 10.18287/1613-0073-2018-2210-219-226.
4. Khryashchev V.V., Priorov A.L., Pavlov V.A., Ivanovsky L.I. Objects segmentation on satellite images using convolution neural networks. Achievements of Modern Radioelectronics. 2019. № 6. P. 28–34. DOI 10.18127/j20700784-201906-04. (in Russian)
5. Plotnikov D.E., Kolbudaev P.A., Loupian E.A. An automatic method for subpixel registration of KMSS-M imagery based on coarse-resolution actualized reference. Computer Optics. 2022. V. 46. № 5. P. 818–827. DOI 10.18287/2412-6179-CO-1098. (in Russian)
6. Vasiliev K.K., Krasheninnikov V.R. Adaptive algorithms for detecting anomalies on a sequence of multidimensional images. Computer Optics. 1995. № 14-15-1. P. 125–132. (in Russian)
7. Buslaev A., Iglovikov V.I., Khvedchenya E., Parinov A., Druzhinin M., Kalinin A.A. Albumentations: Fast and flexible image augmentations. Information. 2020. V. 11. № 2. P. 125. DOI 10.3390/info11020125.
8. Dementyiev V.E., Vasilyiev K.K., Andriyanov N.A. Use of Images Augmentation and Implementation of Doubly Stochastic Models for Improving Accuracy of Recognition Algorithms Based on Convolutional Neural Networks. Systems of Signal Synchronization, Generating and Processing in Telecommunications. 2020. P. 9166000. DOI 10.1109/SYNCHROINFO49631.2020.9166000.
9. Shelkovnikova T.E., Egorov S.F., Gulyaev P.V. Application of generative adversarial neural networks for the formation of databases in scanning tunneling microscopy. Computer Optics. 2023. V. 47. № 2. P. 314–322. DOI 10.18287/2412-6179-CO-1144. (in Russian)
10. Zhai J., Zhang S., Chen J., He Q. Autoencoder and Its Various Variants. IEEE International Conference on Systems, Man, and Cybernetics. 2018. P. 415–419. DOI 10.1109/SMC.2018.00080.
11. Pol A.A., Berger V., Germain C., Cerminara G., Pierini M. Anomaly Detection with Conditional Variational Autoencoders. 18th IEEE International Conference on Machine Learning and Applications. 2019. P. 1651–1657. DOI 10.1109/ICMLA.2019.00270.
12. Radford A., Kim J.W., Hallacy C., Ramesh A., Goh G., Agarwal S., Sastry G., Askell A., Mishkin P., Clark J., Krueger G., Sutskever I. Learning Transferable Visual Models from Natural Language Supervision. [Electronic resource] – Access mode: https://arxiv.org/pdf/ 2103.00020, date of reference 13.01.2024.
13. Gonog L., Zhou Y. A Review: Generative Adversarial Networks. 14th IEEE Conference on Industrial Electronics and Applications. 2019. P. 505–510. DOI 10.1109/ICIEA.2019.8833686.
14. Ahmad A., Andriyanov N.A., Soloviev V.I., Solomatin D.A. Application of deep learning for augmentation and generation of an underwater data set with industrial facilities. Bulletin of the South Ural State University. Ser. Computer Technologies, Automatic Control, Radio Electronics. 2023. V. 23. № 2. P. 5–16. DOI 10.14529/ctcr230201. (in Russian)
15. Rombach R., Blattmann A., Lorenz D., Esser P., Ommer B. High-Resolution Image Synthesis with Latent Diffusion Models. IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022. P. 10674–10685. DOI 10.1109/CVPR52688.2022.01042.
16. Andriyanov N. Using Neural Networks to Identify Parameters of Autoregressive Model with Multiple Roots of Characteristic Equations. 6th IEEE International Conference on Information Technology and Nanotechnology. 2020. P. 9253351. DOI 10.1109/ITNT49337.2020. 9253351.
17. Andriyanov N.A., Dementiev V.E., Tashlinskiy A.G. Detection of objects in the images: from likelihood relationships towards scalable and efficient neural networks. Computer Optics. 2022. V. 46. № 1. P. 139–159. DOI 10.18287/2412-6179-CO-922. (in Russian)