Determination of quality metrics of elaziğ cherry marble using image processing and artificial intelligence

This thesis aims to model and analyze the quality classification using artificial intelligence assisted image processing techniques by determining scientific quality metrics to increase the economic value of the Elazığ Cherry Marble, mined exclusively in the Alacakaya district of Elazığ. Currently, quality assessment processes for natural stones are largely conducted using subjective methods based on expert opinions, resulting in costly and error prone results. Therefore, texture, color, and vein patterns were analyzed on the generated high resolution image dataset, and quality metrics were determined based on expert opinions. The images were processed using deep learning architectures, categorizing marble surfaces into main classes (A, B, and C) and subclasses (A1, A2, and A3). Classification accuracy was achieved with ResNet50 and Support Vector Machines (SVM) with 95.80% accuracy. Pattern comparisons were performed to ensure visual consistency between slabs from the same rock, and block affiliation was determined with 44-92% accuracy using the Local Binary Patterns (LBP) method. To increase classification accuracy and deliver realistic images suitable for e-commerce applications, super resolution techniques were applied. By refining vein details using the RSMAN model, the classification accuracy rate was increased to 96.4%. Furthermore, surface images were evaluated using classical and contemporary art approaches and comparisons were made between artificial intelligence and visual art expert assessments. Ultimately, this thesis presents a holistic approach that automates the quality assessment of Elazığ Cherry Marble based on concrete metrics, while simultaneously combining technical accuracy and aesthetic sensitivity.