支持向量机在炉膛火焰监测中的应用

提出一种利用支持向量机算法监测炉膛火焰燃烧状态的方法。通过对几种监测方法的对比发现,对于火焰图像,基本都是从火焰亮度(通过数字图像处理的方法将其分解为R、G、B三原色图案)、火焰面积(摄像机录像的燃烧火焰像素面积)等几个方面提取数据,来分析并最终得到火焰状态的监测结果。在分析支持向量机算法原理的基础上,利用MATLAB语言编写应用程序来实现对图像原始数据的分类分析。在炉膛火焰强度监测中,首先运用数字图像处理的方法获取炉膛火焰燃烧的原始数据(火焰亮度的R、G、B三原色数据),然后利用SVM算法找到图像数据分类最佳的平面(最大间隔超平面),对原始图像数据进行辨识分类,以获得火焰强度监测结果。通过对炉膛火焰原始数据的分析,提取完全燃烧图像的数据点,与火焰图像总的燃烧像素点进行对比,可以得到较准确的燃烧效果分析图,即把结果分为优(完全充分燃烧)、中(大部分充分燃烧)、坏(少部分充分燃烧)三类,然后通过对分类结果的观察分析来采取相应的措施。仿真实验结果表明,利用这种方法分析火焰燃烧图像,可以有效地、实时地判断出火焰的亮度及燃烧状态并准确的分析其燃烧状态。

Application of Support Vector Machine in the Furnace Flame Monitoring

A method of using support vector machines to monitor the state of the furnace flame burning was introduced.By comparing of several methods of monitoring,the state of flame monitoring results was analyzed by the dates which was the basic characteristics of flame bright（it would be broken down into its R,G,B pattern by digital image processing methods）,the flame area（the size of the flame burning of the video camera pixel） and other aspects of data extraction.Based on the principle of support vector machine（SVM）,the original image data in MATLAB was analyzed.In the furnace to monitor the intensity of the flames,first of all,digital image processing methods was used to obtain the raw data of furnace flame burning（brightness of the flame RGB date）,and SVM was used to find the best plane（the largest interval hyper-plane） of image date classification algorithm,the original image data was identified categories in order to gain the strength to monitor the results of the flame.By analyzing the raw data of the furnace flame,incomplete combustion of the image data was extracted.With the flames of the burning of the total image pixels compared to a more accurate analysis of the effect of burning map,that was,the results were divided into good（which was fully completely combustion）,medium（which was mostly full combustion） and bad（which was a small number of full combustion）,and then by observing and analyzing the results of the classification to take corresponding measures.The simulation results show that in this way the image analysis of the flame burning is an effective way to real-time judge of the flame＇s brightness and combustion state and accurate analysis of the state of combustion.