Research On The Flame Recognition And Feature Parameter Extraction Based On Image Processing

In modern society,the fire is still the major disasters that threaten people lives and property, the disadvantage in sensitivity and reliability of the traditional smoke and temperature type fire detection methods made them hard to satisfy the need of modern fire monitoring, this paper proposed a kind of flame recognition method based on the characteristics of the fire flame videoFirst of all,the consecutive frames color image extracted from the flame video on the MTALAB software platform should be on gray processing to reduce the amount of data,and then use the median filtering and Shuangfeng histogram method to realize image filtering and two values processing,use close operation of mathematical morphology to get suspicious regional connectivity, set the appropriate threshold to those domains, given the removal when the area of domain is smaller than the setted threshold, consider the domain as the the bright spot noise, take the frame difference method to achieve the dynamic target tracking of the rest connected region Based on the analysis of the difference of dynamic and static characteristics of flame and the interference region,we find that the frame of flame area changes, the flicker frequency, flame regions of the cusp number, centroid distance and interference region are obviously different, so this paper chooses the four characteristics of flame characteristics as a criterion of flame recognition.at last,the BP neural network is trained through a large ammount of the four kind of characteristic parameters of flame area and the interference areas,After training, the network can identify the flame image from the interference region, and the recognition results of new flame image can prove the the availability and accuracy of the algorithm.The experimental results show that the flame detection algorithm proposed in this paper can realize the basic monitoring of early fire alarm.