动态火焰图像分割及在电熔镁炉视频监控中的应用

火焰的颜色、亮度分布、形态变化规律在不同工况下均不相同，表现出多模态的特点;此外，背景环境的反光、气雾、粉尘等干扰使得目标区域和背景高度耦合，因此传统分割算法不能确保在多种工况下的准确性.本文提出将多元图像分析和人工经验相结合的火焰图像分割方法，通过主成分分析方法对图像降维、构造得分柱状图，进而分割出炉口火焰区域.将本文方法应用于电熔镁炉生产的视频监控中，结果验证了该方法的性能.

A Dynamic Flame Image Segmentation Method and Its Application in Video Monitoring of Fused Magnesium Furnace Process

The color， brightness distribution and shape pattern of flames vary under different working conditions， thus showing multi-modality. In addition， the background environment is highly coupled with the foreground image because of the reflection， aerosol， dust and other interferences. Traditional segmentation algorithms cannot ensure accuracy under a variety of conditions. A method of flame image segmentation based on multivariate image analysis and expert knowledge is proposed. The principal component analysis method was used to reduce the dimension of images and to construct the scatter plot of the scores， so as to obtain the regions of flame segmentation. The method was applied to the video monitoring of the production of fused magnesium furnace， which verified the performance of this method.