Machine Learning Fire Visual Recognition Based on Fusion of SFF-SVM

A video flame detection algorithm based on the fusion of SFF-SVM (Six Feature Fusion and Support Vector Machine) is proposed for the common phenomenon of low detection rate and high false alarm rate in current fire detection methods. Grayscale equalization, image binarization, and median filtering are used to obtain and process the flame image in the video, and the flame color model is combined to separate the flame suspected area; the flame roundness, flame sharp corner features, flame texture, area change rate, color features and overall mobility features of the suspected flame area are utilized to fuse the feature data to improve the flame recognition rate and accuracy, and based on the established flame dataset, the flame recognition rate and accuracy are improved. Training and testing with an 8:2 ratio, the recognition results on the test set show that the model recognition accuracy reaches 92.8%.