Automated System For Classifying Images of Video Streams

The purpose of research is timely detection of a fire center in the stage of its development can reduce both material and human losses. Therefore, the development of models, methods and algorithms for managing the monitoring system of fire and medical and environmental safety, ensuring an increase in its efficiency through the analysis of video data from unmanned aerial vehicles, is an urgent task.

Methods. The method of classifying aerial photographs of a video sequence when monitoring a fire situation in an autonomous territorial unit assumes their segmentation into identical rectangular segments of a given size and assigning them to one of three classes: smoke, flame, indifferent class. The “strong" and “weak" classifiers are used to classify the segments. The Walsh-Hadamard transform was used to form descriptors for "weak" classifiers. Descriptors are calculated for three "weak" classifiers. First, the Walsh-Hadamard transform is calculated for the window of the entire segment and its spectral coefficients are used for the first "weak" classifier. Then descriptors are calculated for two windows, the sizes of which are two and four times smaller than the sizes of the original window.

Results. Timely detection of a fire in the stage of its development can reduce both material and human losses. Therefore, the development of models, methods and algorithms for managing the fire and medical-environmental safety monitoring system, providing an increase in its efficiency through the analysis of video data from unmanned aerial vehicles, is an urgent task. The classifier consists of three independently trained neural networks - "weak" classifiers. To combine the outputs of neural networks, a simple ensemble averaging block is used. Software for classifying aerial images has been developed, which makes it possible to form a database of segments of "smoke" and "flame" classes, determine the twodimensional Walsh-Hadamard spectrum of aerial image segments, train fully connected neural networks and conduct exploratory analysis to study the relevance of two-dimensional spectral coefficients.

Conclusion. When conducting experimental studies on video sequences containing flame and smoke, the average value of smoke detection accuracy was 86%, flame - 89.5%. False positives for smoke detection averaged 13% and for flame detection 4.5%. To configure and validate the classifiers, we used real data from CCTV cameras in open spaces

1. Anna V. Pyataeva and Oleg E. Bandeev. Video Based Flame and Smoke Detection. Journal of Siberian Federal University. Engineering & Technologies, 2019, vol. 12(5), pp. 542-554.

2. Tomakova R. A., Emelyanov S. G., Philist S. A. Intellektual'nye tekhnologii segmentatsii i klassifikatsii biomeditsinskikh izobrazhenii [Intelligent technologies for segmentation and classification of biomedical images]. Kursk, Southwest State University Publ., 2012. 222 p.

3. Dyudin M. V., Zuev I. V. Chudinov S. M., eds. Avtomatizirovannaya sistema dlya klassifikatsii zabolevanii molochnoi zhelezy po rentgenovskim mammograficheskim snimkam [Automatic classifiers of complexly structured images based on multi-method technologies of multicriteria choice]. Voprosy radioelektroniki. Seriya: Sistemy i sredstva otobrazheniya informatsii i upravleniya spetstekhnikoi (SOIU) = Problems of Radio Electronics. Series: Systems and Means of Information Display and Control of Special Equipment (SOIU), 2015, is. 1, рр. 130-141.

4. Dabagov A. R., Malyutina I. A., Kondrashov D. S., eds. Gibridnaya neironnaya set' s makrosloyami dlya meditsinskikh prilozhenii [Automated system for the classification of breast diseases by x-ray mammographic images]. Neirokomp'yutery. Razrabotka, primenenie = Caspian Journal: Management and High Technologies, 2019, no. 4 (48), pp. 10-24.

5. Philist S. A., Shatalova O. V., Efremov M. A. Gibridnaya neironnaya set' s mak- rosloyami dlya meditsinskikh prilozhenii [Hybrid neural network with macro layers for medical applications]. Neirokomp'yutery. Razrabotka, primenenie = Neurocomputers. Development, Application, 2014, no. 6, pp. 35-39.

6. Dabagov A. R., Gorbunov V. A., Filist S. A., Malyutina I. A., Kondrashov D. S. An Automated System for Classification of Radiographs of the Breast. Biomedical Engineering, 2020, vol. 53, no. 6, pp. 425-428. https://doi.org/10.1007%2Fs10527-020-09957-7

7. Shatalova O. V., Kuzmin A. A., Kasim K. D. A., eds. Metod klassifikatsii slozhnostrukturiruemykh izobrazhenii na osnove samoorganizuyushchikhsya neirosetevykh struktur [Method of classification of complex structured images based on self-organizing neural network structures]. Radiopromyshlennost' = Radio Industry, 2016, no. 4, pp. 57-65.

8. Filist S. A., Tomakova R. A., Degtyarev S. V., Rybochkin A. F. Hybrid Intelligent Models for Chest X-Ray Image Segmentation. Biomedical Engineering, 2018, vol. 51, no. 5, pp. 358-363. https://doi.org/article/10.1007/s10527-018-9748-5

9. Dabagov A. R., Malyutina I. A., Kondrashov D. S., eds. Mnogosloinye morfologiches- kie operatory dlya segmentatsii slozhnostruk-turiruemykh rastrovykh polutonovykh izobrazhenii [Multilayer morphological operators for segmentation of complex structured raster halftone images]. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Seriya: Upravlenie, vychislitel'naya tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the Southwestern State University. Series: Management, Computer Engineering, Computer Science. Medical Instrumentation, 2019, vol. 9, no. 3, pp. 44-63.

10. Filist S. A., Dabagov A. R., Tomakova R. A., Malyutina I. A., Kondrashov D. S. Metod kaskadnoi segmentatsii rentgenogramm molochnoi zhelezy [Method of cascade segmentation of breast radiographs]. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Seriya: Upravlenie, vychislitel'naya tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the Southwestern State University. Series: Management, Computer Engineering, Computer Science. Medical Instrumentation, 2019, vol. 9, no. 1 (30), pp. 49-61.

11. Malyutina I. A., Kuzmin A. A., Shatalova O. V. Metody i algoritmy analiza rentgenogramm grudnoi kletki, ispol'zuyushchie lokal'nye okna v zadachakh obnaruzheniya patologii [Methods and algorithms for analyzing chest radiographs using local windows in pathology detection tasks]. Prikaspiiskii zhurnal: upravlenie i vysokie tekhnologii = Caspian Journal: Management and High Technologies, 2017, no. 3(39), pp. 131-138.

12. Kudryavtsev P. S., Kuzmin A. A., Savinov D. Yu., Filist S. A., Shatalova O. V. Modelirovanie morfologicheskikh obrazovanii na rentgenogrammakh grudnoi kletki v intellektual'nykh diagnosticheskikh sistemakh meditsinskogo naznacheniya [Modeling of morphological formations on chest radiographs in intelligent diagnostic medical systems]. Prikaspiiskii zhurnal: upravlenie i vysokie tekhnologii = Caspian Journal: Management and High Technologies, 2017, no. 3 (39), pp. 109-120.

13. Kassim K. D. A., Kuzmin A. A., Shatalova O. V., eds. Formirovanie priznakovogo prostranstva dlya zadach klassifikatsii slozh-nostrukturiruemykh izobrazhenii na osnove spektral'nykh okon i neirosetevykh struktur [Formation of a feature space for classification problems of complex structured images based on spectral windows and neural network structures]. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta = Proceedings of the Southwest State University, 2016, no. 4(67), pp. 56-68.

14. Tomakova R. A., Filist S. A., Durakov I. V. Programmnoe obespechenie avtomaticheskoi klassifikatsii rentgenogramm grudnoi kletki na osnove gibridnykh klassifikatorov [Software for automatic classification of chest radiographs based on hybrid classifiers]. Ekologiya cheloveka = Human Ecology, 2018, no. 6, pp. 59-64.

15. Bilkent database. Available at: http://signal.ee.bilkent.edu.tr/VisiFire/Demo/Fire- Clips/. (accessed 10.09.2021)