Research On The Method Of Detecting The Flame Area Of Forest Fires

Forests are precious natural resources of human beings.Forest fires often broke out suddenly and they can spread fast,so they are difficult to extinguish.Therefore,forest fires are the most harmful to the forests and their surroundings,and monitoring forest fires is very important.The technology to monitor forest fires based on videos and images has been developed rapidly and it is widely used in recent years.Flame is the main visual feature of forest fires.Therefore,from the point of view of fine-grained,in order to improve the accuracy of detecting flame pixels,a new method for detecting the flame area in forest fire images with complex environments,such as withered grass and light is studied.The flame area detected can be used to judge whether there is a fire or not,and it also provides more accurate flame area for predicting development and trend of the forest fire,so it can provide strong technical support for fire fighting.The color of flame pixels is usually between red and yellow,and flame is flickering,so the current methods of extracting the flame area are mostly based on the color feature and motion feature of flame.However,there is often a strong wind in the wild forest,so some non-flame objects which conform to the color model of flame are misjudged as flame due to the wind blowing them.In addition,when it is not windy,the flame pixels which burn stably do not flicker obviously.Therefore,when a certain area in the image is flame for a short time,there will be missed detection if we use the method of detecting moving targets.This missed detection often occurs inside the flame area,and it especially occurs in the near-white centra flame.Therefore,this paper studies the method of detecting the flame area based on flame color feature and texture feature.The main contents of this paper are as follows.Firstly,the suspected flame area is extracted.The fiercely burning flame under the bright sunlight shows a ring-shaped color change from white to yellow to red which is from inside to outside.Then there will be the near-white centra flame,which is obviously different from the color model of flame which has been built,resulting in incomplete segmentation of the flame area.In this paper,an effective and universal color model of flame is built to solve the problem of missing near-white centra flame,then it can separate the flame area from the background whose color is different from the flame,so the suspected flame area can be obtained.Secondly,the texture feature of flame is extracted.The image with the suspected flame area is cut into image blocks with a certain size.In this paper,the texture feature of each image block is extracted by using non-downsampling wavelet transform and gray level cooccurrence matrix,and this method of extracting texture feature is compared with other methods.The results show that the method of extracting texture feature in this paper is more effective.Then the image blocks are classified to get the preliminary flame area.In this paper,a model for recogniting flame based on AdaBoost-SVM algorithm is used to classify the image blocks into flame blocks and non-flame blocks,and the preliminary flame area is obtained by retaining the flame blocks.The experimental results show that the recognition rate of AdaBoost-SVMalgorithm is better than SVM.Finally,the preliminary flame area is optimized and the final flame area is obtained.In this paper,an optimization project is proposed to solve the problem of missing detection and false detection of the flame area.Especially,when this optimization project is combined with the color model of flame,the problem of missing detection of the centra flame can be solved.The final experimental results show that the method of extracting the flame area in this paper can extract the flame area more accurately and effectively from various non-flame backgrounds.