Fine Segmentation And Reconstruction Of Black Soil Macropore Based On CT Image

Black soil is recognized as the most fertile resource in the world,and it is also the basis of China’s main commercial grain base.The black soil macropore are the main paths for moisture and nutrient migration,which can provide sufficient nutrients and space for plant crop root growth and microbial activity.Exploring the macropore topological structure is helpful for understanding the occurrence and development of ecological processes in black soil,and also has guiding significance for the improvement of black soil structure and the formulation of reasonable irrigation systems.Due to the complex irregularities and spatiotemporal variability of black soil macropores,limited by the existing image processing methods,the characterization of macropore has the disadvantages of weak automation,low accuracy,and poor robustness.Therefore,accurate segmentation and visualization of macropores is a difficult point in the study of black soil structure.In order to achieve the expression and visualization of the black soil macropore structure,this paper proposes automated and intelligent segmentation and reconstruction methods,which achieve the fine segmentation of pore structure and the accurate description of three-dimensional information,thus ensuring that the segmentation method is adaptive to the changes in the CT image due to factors such as moisture content and physical state.Specifically,the main work are as follows:(1)An adaptive fuzzy C-means for macropore segmentation is proposedIn order to solve the problems of weak automation and low segmentation accuracy of the existing pore segmentation method,this paper proposes an adaptive fuzzy C-means method(AFCM)to segment the pore structure.This method can automatically determine the number of clusters according to the black soil CT image under different physical states,and can realize the flexible expression of the segmentation target in practical applications.The comparison with classical pore segmentation methods(Image J and IPP)proves that the AFCM method has the characteristics of high degree of automationand strong pore segmentation ability.(2)A fuzzy C-means based on grayscale-gradient features for macropore segmentation is proposedAiming at the limitation of AFCM method on the initialization of the membership matrix,a fuzzy C-means method based on grayscale-gradient features(GGFCM)has been proposed.This method can assign the initial membership matrix based on the gray-gradient characteristics of the image.The quantitative and qualitative results show that when the number of iterations of is 10,the segmentation error rate of GGFCM method has been reduced to 9% and 6% of FFCM method and AFCM method,thus proving that the GGFCM method has the highest segmentation accuracy and the fastest computing efficiency.(3)A neighborhood constrained fuzzy C-means method based on grayscale-gradient feature for macropore segmentation is proposedAffected by black soil heterogeneity and partial volume effect,macropores have spatial connectivity and neighborhood similarity.The above two methods only consider individual pixels,resulting in incorrect segmentation of macropore structures.Therefore,this paper introduces eight neighborhood information as constraints to the objective function,and proposes a fuzzy C-means large pore segmentation method based on neighborhood constraints.The experimental results show that the GGFCM-N method has the smallest average area / perimeter relative error(2.98% and 5.46%),distribution entropy(0.8118),inter-class correlation(0.1640),and maximum distribution coefficient(0.1120).These results prove that the GGFCM-N method has a better segmentation performance on pore structures of different shapes.(4)A simplified convolutional network for macropore segmentation is proposedDue to the complex variability and irregularities of black soil macropores,the above three methods all have over-segmented and under-segmented pore structures.And considering that the segmentation method should be robust to black soil CT images under different physical conditions,this paper proposes a simplified convolutional network(SCN)to automatically segment macropores.This method can extract the multi-dimensional features such as the shape,texture,and color of the macropore structure through multiple convolution learning methods.In practical applications,it can comprehensively express the macropore features from multiple angles.The experimental results showthat the SCN method has the highest segmentation accuracy rate(99.82%),segmentation accuracy(99.61%)and segmentation recall rate(99.93%),proving that the SCN method has the best macropore segmentation performance and the strongest robustness for different types of black soil CT images.(5)A parallel thinning method for constructing the macropore skeleton model is proposedThe three-dimensional characteristics of macropores are an important condition for evaluating the function of black soil.Therefore,to realize the visual display of the three-dimensional spatial structure of macropore and the full expression of three-dimensional topological features of macropore,this paper,this paper explores the reconstruction method of three-dimensional macropore models.Based on the pore binary image,the applicability of the ray casting method to the visualization of the three-dimensional spatial structure of black soil macropore was verified.Meanwhile,it is proved that the macropore skeleton model extracted by the parallel thinning method has good connectivity,refinement and centrality,and can accurately describe the topological characteristics of black soil macropores.