To improve performance in the state identification of boiler combustion flame images

a state identification method based on the Log-Gabor wavelet and kernel principal component analysis (KPCA) with fractional power polynomial models is proposed. Flame images are filtered by the log-Gabor filter bank. The texture feature vectors of the images are constructed with the use of the mean and standard deviation of the filtered image. KPCA with fractional power polynomial models is utilized to reduce the dimension of the texture feature vectors. These dimension-reduced texture feature vectors are classified by a support vector machine. Experiment results show that the proposed method can accurately extract the texture features of the flame images. Compared with the feature extraction method based on the Log-Gabor wavelet and two other feature extraction methods based on the Gabor wavelet

the proposed method has a higher classification rate of 76%. The variance proportion of the first principal component increases as the kernel parameter increases. A state-identification method of boiler combustion flame images is proposed in this study. High classification accuracy can be achieved through a reduction in the dimension of the texture feature vectors of flame images. Experiment results show that the proposed method can obtain high classification accuracy. It also exhibits a short running time and good real-time performance.