Robust Flame Detection Based on YOLOv5 and Bidirectional Feature Pyramid

Fire is a great threat to people's lives and property, and its casualties are in a high position in all kinds of disasters. So it is of great significance to find and prevent fires. With the rapid popularization of hardware equipment and the progress of AI technology, the fire detection technology based on deep learning is worth exploring and studying in dealing with various fire scenarios. This paper proposes a fire detection algorithm based on YoloV5 and feature pyramid. In order to solve the problem that the network layer is too deep and the gradient disappears, we first introduce the residual structure Resnet network. Secondly, in order to further improve the semantic expression ability and positioning, the neck adopts the FPN+FAN structure, which can complement multi-scale fusion along the top-down and bottom-up paths. Select 5682 and 2338 images to train and test the model. Through practice, it can be concluded that this method can solve this problem and has good robustness to different lighting conditions and different fire conditions.