Fire Flame Recognition Algorithm Based On Support Vector Machine

The hazard of fire for the people is well known. Thus, how to explore more reliable, andtimely fire detection system is a topic of concern. Fire flame recognition method method base ondigital image processing is a non-traditional early fire flame recognition method. Fire flamerecognition method based on Support Vector Machine and digital image processing is studied inthis paper.This paper presents a fire flame image segmentation algorithm based on the HSI colormodel, and extracts a more complete suspected fire flame area for pretreatment. According tosome of the information characteristics of the early flame, the features of fire flame are extracted,including shape of the fire flame, color of the fire flame, and texture of the fire flame. Thesefeatures, as a criterion of identification of early fire flame and be able to distinguish the rest ofthe interference.Finally, we discuss the use of support vector machine to identify the fire flame image andmake use of the flame image samples and disruptors image to experiment. As the support vectormachine classification and recognition performance greatly depends on the selection of thekernel parameter, this article focuses on the selection of the kernel parameter. This paperanalyzes the impact of the model established by the grid search method, PSO algorithm and GAalgorithm to optimize the kernel parameter for the rate of fire image recognition. In addition, thispaper presents an improved particle swarm optimization for parameter optimization in supportvector machine. Experimental results show that the improved algorithm optimizationperformance is better, able to effectively identify fire flame image, increasing the accurateprediction rate of fire, to further reduce the false alarm of fire detection systems.