摘要：One of the most common and dangerous natural disasters in human production and life is fire. Thus, the application of efficient fire warning technologies to complicated settings is crucial for study. To increase the accuracy and robustness of flame recognition, according to the findings of this research, an intelligent video flame detection system according to multi-feature fusion and double-layer XGBoost might be developed. Firstly, for color segmentation, a new color feature model based on YCbCr and HSV color spaces is created. Then, to filter away static interference objects of a similar hue to the flame, the ViBe algorithm with a new background update method is used, and the candidate is achieved. After that, the skeleton shape feature, Gabor texture feature, growth rate, and centroid change rate are adopted to train the independent XGBoost classifiers, respectively. Finally, the second layer XGBoost algorithm integrates each independent XGBoost classifier in the first layer and outputs flame classification results. The simulated annealing algorithm optimizes the parameters of the double-layer XGBoost model. The suggested technique enhances the detection rate and may be used in a variety of settings, according to the findings of experiments.
