Smoke and fire detection method based on small target data enhancement and balance loss function

The invention relates to a firework detection method based on small target data enhancement and a balance loss function. The method comprises three steps: S1, in a model training stage, a small target data enhancement method SOA is used, and a data enhancement strategy is implemented, so that the firework small target detection capability of a target detection network under various conditions is improved; s2, in a model training stage, a balance loss function BLF is used to adjust loss weights of the smoke and fire small target and a normal target so as to improve detection precision of the smoke and fire small target; and S3, in a reasoning prediction stage, reasoning the to-be-detected image by using the trained smoke and fire detection model to obtain a detection result. The method has the advantages that the detection precision of the small smoke and fire target can be improved, and the dependence on a large amount of labeled data is reduced. 本发明涉及一种基于小目标数据增强和平衡损失函数的烟火检测方法。该方法包括三个步骤：S1：在模型训练阶段，使用基于小目标数据增强