Abstract: Intelligent fire recognition system is an important part of constructing wisdom city and preventing fire, which can well guarantee the safety of people's lives and property. Concerning the contradiction of accuracy and real-time in image fire recognition, a rapid fire image recognition method is proposed. Firstly, the segmentation method combining watershed segmentation and automatic seeded region growing is adopted to make suspected flame region segmentation in a complex environment, and the multithreading technology is used to improve the processing speed, so it is conducive to real-time fire recognition. Then, we extract the significant characteristics such as roundness, sharp corners, corrosion resistance, flame core features such as relative coordinates, the relative area in the suspected areas as fire classification, reduce the dimension of feature space and the amount of calculation. Radial basis function (RBF) neural network is adopted to complete the fire recognition, and it shortens the time of fire recognition and improves fire recognition accuracy. The experiment results show that the suspected area extraction accuracy is 90%, the fire recognition accuracy is 85%, the method can improve the precision and speed of the fire recognition.

Key words: fire recognition, RBF, watershed segmentation, automatic seeded region growing