Research On Early Fire Detection Technology Based On Flame Identification

Fire as one of the important disasters has become one of the serious threats to people’slife and property in the current social production and people’s daily life. Especially in theintensive high-rise buildings, fire’s special characteristics such as spreading quickly, difficultyin fire-extinction and rescue make the technology of fire detection particularly important.However, the traditional fire detection technology with response lag and unreliability can notmeet the needs of modern development, and then the fast and correct judgment of the earlyfire status becomes a hot problem that need the solution urgently. Therefore, the early firedetection technology based on the flame identification is introduced and in-depth research inthis dissertation.The main research content and primary innovation of this dissertation can besummarized as follows:1. Based on the research contributions at home and abroad, the research status of thecurrent fire detection technology is introduced, and then the relevant algorithm’s advantagesand disadvantages of flame identification are mainly described.2. Considering the large amount of flame images data, the use of batch readingtechnology has a distinct advantage in reading image datas. Also, due to the characteristics ofthe image noise has the characteristics of Gaussian distribution, then the characteristics ofGaussian filtering method is researched for smoothing it.3. The complexity of the flame image characteristics is analyzed, then The suspiciousarea segmentation method is presented based on adaptive threshold, and compares with theimage segmentation method that is basis of edge detection and motion detection. The resultsof simulation verify that this method can effectively and accurately segment the suspiciousarea, namely, it can meet the requirements of the flame identification technology.4. It is studied on the basis of the flame image feature parameters, The flame areachange rate, circularity, number of sharp corners and outline change distance as the mainbasis of the flame image identification is extracted by applying the fixed threshold HUTDalgritham.5. According to the extracted eigenvalue, the method of BP(Back Propagation) neural network identification based on multi-criteria is proposed in this dissertation. By thisapproach, the weight and threshold of BP neural network is obtained by training a largeamount of eigenvalues, and then the datas of collected flame image is identified. Comparingwith the method of Bayes classifier and Support vector machine classifier, the presentedapproach can not only quickly and accurately identify the flame characteristics, but alsoimprove the accuracy of flame identification, and also an improved discriminant method isprovided for early fire detection.