Research On Algorithm Of Forest Smoke And Flame Detection Based On Image Features

Most traditional forest fire image detection algorithms only analyze smokeor flame feature to finish detecting. However, these detection results are easilyaffected by illumination and chaff in forest. So the accuracy of the detection islow.As a result, flame image detection algorithms based on image feature can beimproved in some ways. In fact, early forest fire will produce smoke. The smokeis important signals showing the flame burst. So combining image features ofsmoke and flame is of great significance.This paper researches past flame detection algorithms based on imagefeature. Then combining with color feature, flap feature of the smoke, flame’sarea and the circular degree, this paper designs and implements an algorithm offorest smoke and flame detection based on image features. By experimentalverification, this algorithm finishes detecting the most area of smoke and flamewith less energy consumption. Smoke images’ features contain color feature and flap feature. This paperproposes an detection algorithm based on Sea Computing. First step, change theimage’s color space from RGB to YCbCr. Then, extract the suspected flame area.Second step, with the introduction of probability density function, calculate theprobability density of Cb, Cr component and filtrate the area. Third step, changethe image’s color space from YCbCr to HSI and filter pixels, with theintroduction of similar distance between pixels. Then detect the edge of smokesuspected area, calculating the center of mass of the area. Forth, calculate flapstrength of the area through theory of flap, filter pixels by the threshold. Finishdetecting the presence of smoke in this image.Detect the flame in image existing smoke. First step, process the imagewith gaussian filter, Gabor filter the filter and then with “quality offset”. Thenfilter pixels by the number of pixels in adjacent frame. Second step, calculate thegrowth rate of suspected area by the growth rate formula of flame area. Extractthe edge of the suspected area. Third step, calculate the circular degree of thearea by one formula. Then calculate the correlation coefficient of circular degreewith reference to adjacent frames. Forth step, obtain the frequency of thecorrelation coefficient by Discrete Fourier Transform (DFT). In the end, finishdetecting the flame in order to judge whether there is forest fire or not.The experimental results show that the accuracy of the algorithm of forestsmoke and flame detection based on image features is high. This algorithm issuitable for forest fire warning system.