振荡燃烧火焰图像处理

Image Processing of Combustion Oscillating Flame

应用图像FFT、本征正交分解(POD)和动力学模态分解(DMD)3种方法,对在振荡燃烧环境下的火焰图像进行分解,得到火焰脉动幅值、主频以及初始相位.其中通过图像FFT方法分析图像中每一个像素灰度值时间序列,对其进行傅里叶变换,然后得到整体火焰的振荡特性.而POD和DMD方法将每一张火焰图像视为一个样本数据,以向量的形式表达,通过线性代数和矩阵分析等理论对样本数据进行分解.DMD和图像FFT两种方法得到的火焰脉动幅值和主频基本一致,可以得到单一频率(基频和各谐频)脉动特征,而POD方法无法将各种振荡频率对应的脉动结构进行分离,得到总的脉动结构.结果表明上述3种方法均可以捕捉振荡燃烧环境下火焰的主要脉动特征.

Three methods,i.e.,image FFT,proper orthogonal decomposition(POD)and dynamic mode decomposition(DMD),were used to decompose the flame images under the combustion oscillation circumstance,and the fluctuation amplitude,dominant frequency and initial phase were obtained.In the image FFT method,the fast Fourier transform was introduced to analyze the time-varying gray values of each pixel in the images,and then the fluctuating characteristics of the whole flame structure were obtained.The POD and DMD methods treated each flame image as a sample data,which was expressed in the form of a vector.Afterwards,mode decomposition was conducted according to linear algebra and matrix analysis theory.The dominant frequencies and the corresponding amplitudes obtained using the DMD method were in agreement with those obtained using the image FFT method,and the pulsating characteristics of single frequency(e.g.,fundamental and harmonic frequencies)were obtained.The POD method failed to separate the pulsating structures corresponding to the specific dominant frequency,instead,it captured the whole fluctuating structure.The results showed that all the three methods above can capture the primary fluctuating characteristics of flame via images in the oscillating environment.