Burning State Recognition Based On Flame Images Of The Sintering Process Of Rotary Kiln

Flame images can reflect the burning state of the sintering process of rotary kilneffectively. The luminance usually describes the radiation intensity and the burning state inrotary kiln while the flame configuration represents the zone of combustion reaction. Forthe past few decades, the image based burning state recognition of a sintering process hasattracted much attention. Existing approaches mostly follow the pattern using imagesegmentation, feature extraction and pattern classification, which are facing withinevitable challenges, costing an amount of time on image learning and training andprocessing. Meanwhile, some non-structural distortions of flame images caused byexternal factors also trouble the burning state recognition. This thesis develops the burningstate recognition based on flame images and presents two new recognition methods basedon the structural similarity of flame images. The contributions of the paper are two-fold asfollow:1) Based on the structural similarity in the space domain of flame images, a newburning state recognition method is proposed. It can figure out the SSIM from theperspectives of luminance, contrast and structure and recognizes the burning state bycomparing structural similarity of flame images between measured image and givenreferenced images. This method without image training and learning takes an obviousadvantage on simple formula and low computing complexity, which meets the requirementof rapidity of burning state recognition for online rotary kiln control system. A lot ofsimulations show the rapidity and effectively of our method.2) To deal with the small geometric distortions of flame images, an improved indexof wavelet domain structural similarity and a new weighted algorithm of wavelet subband.We present another recognition method on basis of the wavelet domain structuralsimilarity (WDSSIM) of flame image. It compares the similarity between images by thecorrelation of wavelet coefficients in different wavelet subbands and is an extension ofSSIM based recognition method which is not effective enough on flame images withgeometric distortions. Our simulation results demonstrate that this WDSSIM method hasbetter performance on dealing with flame images with small geometric distortion andheavy noise pollution.