Improved Yolov4-tiny for Fire Detection

• Published in: 2024 20th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD) pp. 1 - 5
• Main Authors: Li, Qiyuan, Xiang, Qinghai, Xie, Yu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.07.2024
• Subjects: Climate change; Computational modeling; Disaster management; Ecology; Fire detection; Fire safety; Fires; Intelligent sensors; Machine learning; Object detection; Public security; YOLO; Yolov4-tiny
• DOI: 10.1109/ICNC-FSKD64080.2024.10702272

Fire disasters is one of the most frequent and widespread major disasters that threaten public safety and social development, causing property damage, casualties, ecological damage, social impact, and even climate change. Therefore, how to effectively reduce the losses caused by fire accidents has become an important research topic. Traditional methods based on sensors and computer imaging are used to detect fire accidents, but they have the drawback of having a small sensing area and a high false alarm rate. Image object detection technology is an important research topic in the field of machine learning, and has been widely applied in intelligent video analysis, achieving good application results, especially Yolov4-tiny, which has been widely applied in lightweight object detection. For flame target detection, Yolov4-tiny still has the problems of low accuracy and weights size, which is not conducive to deployment on low computing power devices. This paper improves the Yolov4-tiny model, achieving a high accuracy even with low weights size. The main improvements include modifying the backbone network and adding an output layer. We constructed a flame dataset and conducted experimental verification, and the results showed that the proposed method effectively reduced the weight size and improved accuracy; This is an effective method for deploying flame detection models on low computing power devices.