Research On Fire Detection Technique Based On Image Processing

The fire is frequent and destructive to the people’s life and property, on whichpeople get more and more concerns. For early detection and extinguishment of firedisasters, many new fire detection methods emerge one after another, but they stillcouldn’t meet the higher accuracy and higher recognition rate for fire detection under unknown environments. To solve this problem, a novel SVM algorithm is proposedin this paper.Research on fire detection technology by SVM algorithm mainly involvesextraction of flame features of fire disasters and classification of fire images by SVMalgorithm based on these features. To distinguish from interference source images,this study selects shape, texture and color features of fire flames namely circularity,number of sharp corners, energy, local stationary and red proportion variance etc offire flame images according to contrast between fire flames and candle flames,lighting and other features of interference sources. In terms of classification of fireimages by SVM algorithm, this paper mainly carries out the following aspects ofwork:(1) introduction of the concept Chaos for improvement of PSO(Partide SwarmOptimization) algorithm in order to optimize selection of parameter g of radial basisfunction and penalty factor C; and (2) design of fire detection algorithm based onSVM and proposal of an improved SVM algorithm; and (3) the specific imple-mentation of SVM fire detection algorithm based on Libsvm software package isstudied; and (4) finally proves that fire detection algorithm based on improved SVMalgorithm put forward in this paper has a higher fire recognition rate and better firedetection through comparative experiment with traditional neural network algorithm.The existing problem in the field of fire detection, is solved by SVM algorithm,and objective of this research is achieved with some theoretical and practicalsignificance.