SEMFOGO: An Intelligent Fire Detection System for the Cerrado Biome

• Main Authors: Borges, Natalia O., Fonseca, Livia G. C., Barreto, Priscila Solis, Alchieri, Eduardo A. P., Caetano, Marcos F., Araujo, Daniel C., Resende, Paulo Angelo A., Brandao, Leonardo, Vieira, Lucas
• Format: Conference Proceeding
• Language: English
• Subjects: Cerrado Dataset; Deep Learning; Fire Detection; Forestry; Measurement; Real-time systems; Smoke Dataset; South America; Streaming media; Training; Video surveillance; Wildfire
• ISSN: 2771-5752
• DOI: 10.1109/CLEI60451.2023.10346186

Forest fires have the potential to cause enormous social, economic and, above all, environmental damage. Nowadays, the intelligent and early detection of forest fires is a fundamental technological tool for rescue and emergency agencies in mitigating damage. Among the possibilities, video surveillance technology in conjunction with artificial intelligence and computer vision techniques proves to be an interesting solution. This work proposes the SEMFOGO system, an intelligent system for monitoring and rapid detection of fires in the cerrado region, which currently in South America has an area of 2,045,000 km 2 between Brazil, Bolivia and Paraguay. The SEMFOGO solution uses the massive capture and processing of video streams through a distributed and scalable system and applies a deep learning model that performs a grid classification on parts of the image. This work also presents a new training dataset specific to the cerrado region with smoke contour annotations. The experimental results compared the metrics of several models and show the practical feasibility of the proposed solution for real time fire detection.