Research On Fuzzy Clustering For SOFC Anode Image Segmentation

At present,optical microscopy image analysis technology for characterization of Solid Oxide Fuel Cell(SOFC)anode microstructure is still in a development stage.Three-phase identification is the most significant step from microstructure image processing to its analysis.Due to physical characteristics of composite materials and real imaging environment,there are many uncertainties in porous Ni-YSZ cermet anode micrographs.Fuzzy clustering algorithm can describe the uncertainty of pixel classification by means of membership function,and hence widely applied in image segmentation.Therefore,this thesis deeply investigates existing fuzzy clustering algorithms and develops two effective three-phase identification algorithms for Ni-YSZ anode images.1.Aiming at low precision of SOFC electrode image segmentation,this thesis firstly combines Gaussian Mixture Model(GMM)with fuzzy logic model to propose a novel quantum-inspired mixture clustering model.First of all,a new clique Markov Random Field(MRF)potential function is designed using the average of GMM prior probabilities and fuzzy membership functions of neighborhood pixels,which takes both prior statistical information and spatial neighborhood information of SOFC electrode image into account.In addition,quantum-inspired adaptive fuzzy factor is introduced into the novel MRF clique potential to overcome the problem of manual parameter selection.Lastly,in order to improve the segmentation accuracy,we design the logarithm prior probability function as pixels distance measurement to accurately reflect the distribution characteristics of the data.The experimental results show that the proposed method obtains a higher segmentation accuracy and has a better ability to suppress noise as well as preserve image details.2.Aiming at intensity inhomogeneity and noise in SOFC electrode images,this thesis also proposes a noise-suppressed and bias field corrected fuzzy kernel metric clustering algorithm.According to image statistical model,a principal component analysis based denoising method is firstly used to remove noise in Ni-YSZ anode microstructure images.Secondly,bias field correction theory and kernel metric are introduced into fuzzy clustering algorithm to improve the robustness of three-phase segmentation result to both uneven illumination and noise.Since the following innovations: 1)our de-noising algorithm considers the local similarity in SOFC electrode image,2)we use a set of smooth basis functions linearly represent bias field to ensure its smoothing varying characteristics,3)we utilize kernel metric to replace Euclidean distance in the objective function of fuzzy clustering algorithm,the proposed method can obtain clearer and more accurate segmentation results on SOFC electrode images polluted by both intensity inhomogeneity and noise.This thesis proposes corresponding algorithms based on the above two issues.Experimental results show that the proposed methods are simple and effective,which provides theoretical support for the extraction of SOFC microstructure parameters.