Analysis of ceramic surface defects detected with hybrid optical imaging system using image processing and deep learning techniques

The aim of this study is to develop a high-resolution hybrid system capable of non-contact measurement using a Hybrid Optical Imaging Sensor (HOGS) to precisely detect visible and invisible defects in materials with ceramic and porcelain-like structures and to improve the efficiency of production processes. For this purpose, phase information was used to retrieve data from the holograms recorded with the HOGS system, which combines Lateral Shearing Digital Holographic Microscopy (LSDHM) and Microscopic Fringe Projection Profilometry (MFPP) methods. In order to extract the phase information, commonly used techniques in the literature such as Continuous Wavelet Transform (Mexican Hat, Morlet), Fourier Transform, and Hilbert Transform were preferred. In addition, to further improve the performance of the successful Hilbert Transform, various edge detection techniques including Canny, Prewitt, and Sobel were applied. Finally, the Hilbert+Sobel technique, which made the defect most distinguishable in the images, was used for classification with a deep learning-based ResNet-50 model that was modified accordingly to distinguish defective ceramics from flawless ones. In order to evaluate the impact of the applied image processing methods on classification performance, the raw optical images obtained with HOGS were also classified using the ResNet-50 architecture, and the results were presented comparatively. This study aims to contribute to the reduction of time and cost in the quality control process in ceramic production, while also providing more precise and accurate results than the human eye through remote sensing. Compared to various remote sensing methods used today, this approach offers both cost-effectiveness and fast quality control without damaging the ceramic surface. This study is expected to help overcome the difficulties encountered in surface defect detection in the ceramic and porcelain industry, which operates in many sectors such as construction, electrical-electronics, medicine, art, defense industry, and engineering applications. As the thesis brings together multiple approaches to offer a hybrid system, it has become one of the pioneering works in this field.