Smoke recognition model in an open area using a convolutional and recurrent neural network

Timely detection of the source of ignition is an important issue of protecting people, animals and vast territories from fires. The relevance of this study is due to the fact that the existing visual smoke detection systems have a number of disadvantages that do not allow them to be effectively applied in practice. The surveillance system must rely on visual characteristics and often mistakenly identifies fog and clouds as smoke. The aim of this study is to increase the efficiency of smoke detection by using an advanced smoke detector model based on the "You-Only-Look-Once" neural network and a classifier with long-term short-term memory (LSTM). The main objectives of the study are structural description of the proposed smoke detection model, mathematical description of the trained model and comparative analysis with existing neural network models. By changing the structure of the LSTM, a reduction in the number of layers and cells is achieved, and the performance of the original LSTM is maintained. The proposed method provides a reduction in the number of parameters by several times and a faster processing time on the data set used. The article presents the results of the performance of artificial intelligence systems for a comparative analysis of neural network candidates in the smoke recognition model.

1. Balonin N.A., Sergeev M.B. Besprovodnye personal'nye seti. Sankt-Peterburg: GUAP; 2012. 60 p. (In Russ.).

2. Zajac A.M., Logachev A.A. Informacionnaja sistema monitoringa lesov i lesnyh pozharov s ispol'zovaniem besprovodnyh sensornyh setej. Izvestija Sankt-Peterburgskogo gosudarstvennogo lesotehnicheskogo universiteta im. S.M. Kirova. 2016;3:241–254. (In Russ.).

3. Pjatibratov A.P. Vychislitel'nye mashiny, seti i telekommunikacionnye sistemy. 2-e izd., pererab. i dop. M.: Finansy i statistika; 2004. 508 p. (In Russ)..

4. Hodakov V.E., Zharikova M.V. Lesnye pozhary: metody issledovanija. Herson: ChP Grin' D.S.; 2011. 470 p. (In Russ.).

5. Cao Y., Yang F., Tang Q., Lu X. An Attention Enhanced Bidirectional LSTM for Early Forest Fire Smoke Recognition. IEEE Access. 2019;7:154732–154742. DOI: 10.1109.

6. Cetin A.E., Dimitropoulos K., Gouverneur B., Grammalidis N., Günay O., Habiboglu Y.H., Töreyind B.U., Verstockt S. Video fire detection – review. Digital Signal Processing. 2013;23(6):1827–1843. DOI: 10.1016.

7. Chen Y., Zhang Y., Xin J., Wang G., Mu L., Yi Y., Liu H., Liu D. UAV Image-based Forest Fire Detection Approach Using Convolutional Neural Network. In Proceedings of the 14th IEEE Conference on Industrial Electronics and Applications. 2019:2118–2123. DOI: 10.1109/ICIEA.2019.8833958.

8. Jiao Z. A Deep Learning Based Forest Fire Detection Approach Using UAV and YOLOv3. 1st International Conference on Industrial Artificial Intelligence (IAI). 2019:1–5. DOI: 10.1109/ICIAI.2019.8850815.

9. Hinton, G., Vinyals, O., Dean, J. Distilling the Knowledge in a Neural Network. ArXiv. 2015:1503.02531.

10. Gotthans J., Gotthans T., Marsalek R. Deep convolutional neural network for fire detection. 30th International Conference Radioelektronika (RADIOELEKTRONIKA). 2020;1:1–6. DOI: 10.1109/RADIOELEKTRONIKA49387.2020.9092344.

11. Wu S., Zhang L. Using Popular Object Detection Methods for Real Time Forest Fire Detection. 11th International Symposium on Computational Intelligence and Design (ISCID). 2018;4:280–284.DOI: 10.1109/ISCID.2018.00070.