Abstract: The initial stage of fire is the best time to extinguish the fire, so it has a very important significance for the initial fire detection. The flame area of the initial fire is small and the data samples are few, the traditional machine learning target detection method is difficult to train effectively. In view of the above problems, an image early fire detection system is proposed, and the model based training is studied. The test results show that the model has an accuracy of 97%, the initial fire detection accuracy is high, and the detection speed is fast, and the initial fire can be detected quickly and accurately.

Key words: initial fire detection, target detection, Yolo V5, transfer learning, computer vision