Ivasenko I. Estimation of inhomogeneity and damage of construction materials by methods of structure-adaptive image processing

The dissertation solves an important scientific and technical problem of creating methods and means for computerized assessment of materials and structural elements by creating methods of structure-adaptive image segmentation to quantify their 40 diagnostic parameters. The developed methods include image pre-processing, localization of objects of interest, segmentation, quantitative parameter estimation. It is necessary to highlight the localization of welds and determining the size of welding defects on a complex-structured background, analysis of corrosion damage on images of painted surfaces of arbitrary color, modeling and analysis of laser reinforcement of aluminum alloy with silicon carbide particles, evaluation of porosity of metal coatings, improving the accuracy of tracking motion in electron beam welding, analysis of fractographic images of fractures of heat-resistant steels. A structure-adaptive method of segmentation of weld zones of different shapes on radiographic images has been developed. Due to the introduction of new informative features of defects in the welding area of the oil and gas pipes, a method of defect segmentation was constructed, which increased the probability of correct classification of spherical pores and slags. A method for calculating the depth of defects based on their radiographic images has been developed and the relative errors in calculation the width and depth of defects have been experimentally established. On the basis of a cylindrical color model and a single-scale retinex, a method of structure-adaptive segmentation of corrosion damage on images of painted surfaces of arbitrary color obtained under non uniform illumination was developed. Modelling for different volume concentrations of particles in the surface layer of aluminum alloy based on uniform distribution was performed. This made it possible to compare experimental and simulated images of surface layers cross-section to determine the volume concentration of silicon carbide particles and further prediction the mechanical properties of such surfaces. A method of structure-adaptive localization and segmentation of pores on surface images of oxide ceramic coatings has been developed, which uses the verification of the pore background area, which provides the possibility of calculating the degree of porosity of the coating.