摘要：In order to improve the fire detection efficiency of large-scale and complex environment, the paper built a model of flame detection on visual selective attention method. In this model, saliency regions are obtained by Fourier spatial and temporal spectral residual, saliency objects are generated by using a object detection method based on the weight of the color space clustering, and saliency points are found by difference block inverse probability .According to the rules of flame’s color we can judge the risk of fire. The experimental results show that the flame detecting method can find different flames in wild scenes ,and it is very accurate, effective and robust.
