Research On Fuzzy C-Means Clustering Algorithm For Image Segmentation And Its Application In Oil Reservoir Displacement

In recent years,with the rapid development of our country’s economy,oil demand has also been rising year by year.By 2020,our country’s oil import dependency rates has reached 73%,which will seriously restrict the sustained and stable development of our country’s economy.In order to ensure energy security,it is necessary to improve further the production efficiency of our country’s oil reservoir and alleviate our country’s dependence on import oil.At present,low permeability oil reservoir resources have become the main battlefield of our country’s oil exploitation.However,the type of reservoir usually faces problems,such as low reservoir permeability,low abundance,and low single-well productivity,resulting in low displacement efficiency.To further explore the displacement process,the thesis applies fuzzy c-means clustering algorithm to water-oil displacement image,but it still faces some problems such as high computational complexity,many human interventions,and low segmentation accuracy.The thesis takes the research breakthrough point by the problems,and introduces strategies such as histogram information,local spatial features,regularization constraints,entropy theory,and superpixels into the fuzzy c-means clustering algorithm to improve the recognition ability of water and oil pixels from multiple angles.The accurate identification of water and oil not only provides theoretical guidance for the exploitation of low permeability oil reservoirs,but also indirectly promotes the integration of intelligent technology and the real economy.The main contributions are summarized as follows:(1)Fast and robust fuzzy c-means based on morphological reconstruction and membership filtering(FRFCM)is proposed.Firstly,utilizing the morphological reconstruction filtering to suppress noise interference.Secondly,completing the pixel clustering on the histogram of the filtered image.Finally,adopting membership filtering to modify the pixel category.So FRFCM not only can integrate fully the local spatial information of the image,but also relieves the computational complexity.Compared with state-of-the-art algorithms,FRFCM does not need to repeatedly calculate the distance between the neighboring pixels and the cluster center.It only needs to embed filtering strategies before and after clustering to exhibit local correlation.The design can take the advantage of the number of gray levels which is less than the num-ber of pixels,simultaneously,can weaken the dependence of the filter on the noise type.Experiments show that FRFCM has better robustness than other algorithms to synthetic images with different types and levels of noise,and obtains satisfactory visual effects in medical images,aurora images,and benchmark datasets.Their corresponding quantitative indicators also further verify the effectiveness of FRFCM.In addition,FRFCM has excellent applicability in water-oil displacement images.It highlights superiority of FRFCM from both visual effects and quantitative indicators.The proposed FRFCM can provide a technical guarantee for the improvement and application of water-oil displacement theory.(2)Fuzzy clustering algorithm based on robust self-sparseness(RSSFCA)is proposed.RSSFCA utilizes regularization under Gaussian metric to obtain proper sparse memberships that can weaken noise interference.Then RSSFCA employs connected-component filtering algorithm based on area density balance strategy to merge the isolated regions,and reduce over-segmentation.The two contributions promote synergistically the segmentation effect of RSSFCA.Compared with state-of-the-art algorithms,RSSFCA adopts a self-optimization strategy to overcome the non-homogenous interference.At the same time,RSSFCA selects the adaptive region merging scheme to replace spatial constraints.The composition not only exerts anti-noise robustness of sparse optimization,but also avoids the repeated calculation of neighborhood information,which can improve further the generalization of RSSFCA.Experiments show that RSSFCA has a good applicability in synthetic images with significant noise interference,and can obtain ideal segmentation results in public datasets.Their quantitative indicators also verify further the effectiveness of RSSFCA.In addition,RSSFCA can also accurately identify the oil and water components in the water-oil displacement image,which can lay a theoretical foundation for the calculation of oil displacement efficiency,the analysis of pore fluid characteristics,and the exploration of seepage laws.(3)Superpixel-basd fast fuzzy c-means(SFFCM)is proposed.Firstly,utilizing adaptive morphology with multi-scale information to optimize gradient image.Secondly,employing watershed transformation to obtain superpixels with accurate contours.Finally,using color histogram to complete the pixel clustering.The strategy not only avoids the limitation of the rule window on the neighborhood information,but also greatly improves the segmentation accuracy of SFFCM.Compared with state-of-the-art algorithms,SFFCM introduces adaptive area attributes,which can effectively compress the redundant pixels of the image while building a broad neighborhood field.The strategy can not only improve calculation efficiency,but also improve visual effect.Experiments show that SFFCM has a good robustness to synthetic images with different types and levels of noise,maintains good segmentation effects in public datasets.Their quantitative indicators also verify the generalization of SFFCM.In addition,for highresolution water-oil displacement images,SFFCM can avoid pixel-by-pixel calculations under the action of superpixels and quickly completes water-oil segmentation without losing the original data.SFFCM is convenient for hardware deployment and can provide technical support for in-depth study of oil displacement mechanism.(4)Automatic fuzzy c-means clustering framework based on density peak(AFCF)is proposed.Firstly,utilizing the superpixel algorithm to obtain the presegmentation.Then employing density peak algorithm to build a decision graph.Next,using the density balance strategy to realize the automatic evaluation of the number of clusters.Finally,adopting fast fuzzy clustering algorithm based on prior entropy to complete image segmentation.AFCF not only avoids the artificial setting of the number of clusters,but also improves the final segmentation results.Compared with state-of-the-art algorithms,AFCF can achieves fully automatic image segmentation and overcomes human intervention.On one hand,AFCF expands the application of the density peak algorithm in image segmentation.On the other hand,AFCF takes the advantages of superpixels.The scheme can balance better the calculation efficiency and segmentation accuracy of AFCF,and clarify the direction for the cross fusion of different algorithms.Experiments show that AFCF can accurately predict the number of clusters of synthetic images with different noise levels,and can complete high-precision image segmentation.Moreover,the stability of AFCF is also verified in the public datasets,and their quantitative indicators prove further the effectiveness of AFCF.In addition,AFCF also performs well on water-oil displacement images,and can accurately identify oil and water pixels,and pave the way for development of displacement theory and the high efficiency of oil reservoir exploitation.(5)Fuzzy c-means with Student’s t-distribution based on richer spatial combination(FRSC)is proposed.Firstly,utilizing pixel-level spatial information to mine neighborhood features.Secondly,employing regional-level superpixel information to perceive area attributes.Finally,completes image segmentation under the Student’s t-distribution with heavy-tail.The strategy not only integrates the multi-scale structural information of image,but also helps to distinguish image noise and boundary differences.Compared with state-of-the-art algorithms,FRSC constructs a multi-level spatial fusion under the measurement Student’s tdistribution,which overcomes the shortcomings of single spatial information in distinguishing differences.The design retains the advantages of each spatial feature and makes up for their own defects,thereby deeply improving the immunity of FRSC to pixel distribution and noise interference,and ensuring highprecision segmentation results.Experiments show that FRSC has a good robustness to synthetic images with different types and levels of noise,and has also a good segmentation effect on synthetic texture images.Simultaneously,the performance of FRSC is better in public datasets.In addition,in the displacement task,FRSC can still effectively identify the water and oil pixels,which can provide accurate numerical for the research of water-oil tension and reagent viscosity during the displacement process.