Study of Flame Detection based on Improved YOLOv4

Abstract In some complex circumstances, the detection of conflagration mostly depends on smog detectors, which have lots limitations in precision, efficiency and safety. If we make full use object algorithms to detect flame industries, it will benefit people’s safety obviously. Among all kinds algorithms, YOLO series play a very significant role. this paper, propose an improving strategy YOLOv4 enhance its precision based multi-scale feature maps. Firstly, create datasets including almost 4000 high-resolution pictures. Secondly, improvements extraction network are made smaller objects. Finally, total algorithm trained tested our for about 400 epochs. The result show that method can generate high quality great number situations.