Research Of Image Flame Recognition Algorithm Based On HMM-SVM In Spacious Buildings

With the development of computer and image processing technology, image-basedfire detection technology has become a hot research spot in recent years. Comparedwith the traditional fire detection technology, image-based fire detection with fastresponse, wide scope of monitoring, low overhead and other characteristics, is moresuitable for fire detection in Spacious Buildings. This paper analyzes the characteristicsof flame images of image-based fire detection in Spacious Buildings, as well as theinterference factors which are likely to influence. On this basis, image flame recognitionalgorithm based on HMM-SVM in spacious buildings is researched.?1?Flame feature extraction. We have researched the characteristics like circularity,rate of area change, rate of overlapping area, rate of red and green components,correlation coefficient, number of sharp corners, flicker, overall mobility and discretecosine transform in frequency domain. We have also extracted and analyzed the imagefeatures, to prepare for the input of rear recognition models.?2?Flame recognition. The five features with good effect in airspace have beenchosen as the input of Support Vector Machine. Then the dimensionality reduction hasbeen dealt with discrete cosine transform, and the results have been input HiddenMarkov Model. Hidden Markov Model is good at dealing with dynamic signal andSupport Vector Machine has powerful classification ability. On this basis, CombiningHidden Markov Model and Support Vector Machine, a flame recognition algorithmbased on HMM-SVM is researched. Hidden Markov Model is used to calculate thematching values, which are input into Support Vector Machine for second recognition,thus ensuring the accuracy of the recognition results. The experimental verification of flame recognition algorithm based on HMM-SVMhas been done. The results show that in the feature extraction section, circularity, rate ofoverlapping area, rate of red and green components, correlation coefficient, overallmobility have a better performance, and in the flame identification section, comparedwith only Hidden Markov Model or Support Vector Machine, flame image recognitionalgorithm based on HMM-SVM has a higher accuracy rate.