Research Of Flame Recognition Method Based On Multi-Feature Fusion

The image-type fire detection technology is a new detection method, which basedon the machine vision. It will not be subject to the complex factors of the space size orthe bad enviorment. Comparred with conventional fire detectors such as thermaldetectors, smoke detecor and light-sensitive detector, its accuracy and real time morehigh.This thesis first introduces the current development of the traditional fire detectiontechnology, summarizes their drawbacks. Analysis several typical algorithm of flamedetection about the existing image fire detection technology, and discusses the maincomponent of the image-type flame recognition.Collect the fire image information under complex enviroment. Analysising theimage, and using the median filtering and image sharpening to the image pre-processing,eliminating the noise mixe in the image of fire. With the two-tier combination of imagesegmentation algorithm to segmentation the image, in order to get the suspected area ofthe flame. On this basis, analysis the flame characteristics of the suspected region, toextract the flame area change values, round,and the flicker frequency characteristics foridentification in preparation.Finally, using the fuzzy neural network to establish the model of fire detection,Flame characteristics will be extracted as the input, in order to classification of the fireimage. The simulation results show that the algorithm on different scenarios of firedetection with high accuracy.