The Method For The Selection And Recognition Of Fire Flame In The Complex And Large Space

Fire disaster is a kind of devastating addcident that seriously threats the people’s life and property safety, and it always happens suddenly. Therefore,with the rapid development of society and economy, the fire control work is getting more and more important nowadays.The traditional fire detection adopts different sensors to detect wheter there is a fire by analyzing the temperature, spectra, smoke particles and the content of combustible gas in a region. However, it alwyas lead to the failure of fire alarm in large space environment. In order to solve this problem,the video fire detection technology based on on security surveillance system and image processing technology is proposed, and it has a broad and promising application prospect.In this paper,the principle and characteristics of video fire detection technology are summarized.The fire detection method for the complex and large space is mainly discussed. There are three parts in this paper. Firstly,the fire video obtained by cameras is processed and seperated to form images frame sequence.And then,the noise in images is filtered and the fire suspicious region is extracted by putting the segmentation technology into use.The ratio of component area of the red to that of the green,the circularity,the amount of sharp angle,the variance rate of area,the related coefficient,the similarity and the whole mobile characteristics in the flame suspected region are analyzed and calculated.Finally,two kinds of attribute reduction algorithms of the rough set are applied to reduce and recognise the dimension of characteristic vector which is formed by flame criterion,to remove the redundant information, and to lower the sample dimension.In the RS-SVM classifier,the simplified characteristic vetor will be classified and judged.In this paper, a large number of simulation experiments for the recognition of flame image in the complex space are conducted.The results illustrated that the redundancy property of sample sets can be greatly and effectively eliminated when the attribute reduction algorithm of rough set theory is introduced into the support vectormachine classifier. Based on the rough set as the prefixion system, the training speed of support vector machine and the recognition speed of classifier are improved with the guarantee of fire recognition accuracy. Therefore, the algorithm has a high recognition rate and low false rate.In fact, this sector will be especially promising and highly demanded in the near future.