Flame Detection Technology Based On Alpha Channel Of Video Image

With the rapid development of the industry of our country, the requirements of each link of the industrial production become stricter and stricter gradually. As the main facility used for the treatment of industrial exhaust, the safety, reliability and environmental protection property of elevated torch or ground torch system become key points to which petrochemical enterprises pay much attention. Therefore, the flame detection technology based on image processing has important practical significance to the detection and diagnosis of the combustion status of torch system.The deficiency of the current flame detection algorithm is analyzed in this article, while with the combination with the practical applications given by Liaoning Province Electronic Information Research Institute who is our partner for this project, a new flame description method, which is combined with the information of color, status and penetrability of the flame, is proposed, and the detection of flame is carried based on the method, so that the intelligent diagnosis of combustion status is realized. The main work of the article includes the following parts:First, the priors of the color and form of flame are used for judging whether flame exists in the image. The foreground area of the flame is located through MeanShift algorithm which is based on HSV Color Space. The form information of the flame is obtained through Canny edge detection algorithm. Because threshold value needs to be given in advance in the traditional Canny algorithm, and selection of the threshold value will directly influence the result of foreground image segmentation, a self-adaptive Canny algorithm is realized in this thesis, and the experiments showed that the algorithm has ideal effect.Second, a combustion status distinguishing method which is based on alpha channel image is proposed in this article. In this method, the foreground image of the flame, i.e. alpha channel image, is extracted through a guided filter. Because a lot of user interaction is needed for the guided filter, and the human machine interaction cannot realize the frame-by-frame input of the user information during the processes of flame image detection and recognition, we propose an algorithm which combines foreground demarcation with region growing, in order to offer reliable data statistic information for subsequent combustion status diagnosis.Third, an analysis mechanism of flame combustion status is developed in this work. The main task of the analysis mechanism includes the following part:(1) characteristic parameters and calculation methods thereof which are used for characterizing the combustion status are defined; (2) characteristic parameters of several flame images are extracted and counted; (3) corresponding relationships of the characteristic parameters and actual combustion status are analyzed through experiments. According to the simulation experiment results, the multiply characteristic parameters defined in the thesis can truly reflect the change of combustion status.Last, we study relevant theories of support vector machine, and classification tests for the data of this thesis are carried out through LIBSVM environment, while good experimental results are achieved. So that the technology can meet the requirements of safety, economical efficiency and environmental protection property of torch system operation, and is helpful for workers to accurately judge the actual working conditions of the combustion of torch system. Therefore, the flame detection technology based on the alpha channel of video image has feasibility and effectiveness.