Sarı, FilizUlaş, Ali Burak2022-06-232022-06-2320220765-00191958-5608https:/dx.doi.org/10.18280/ts.390238https://hdl.handle.net/20.500.12451/9468Manually detecting defects on the surfaces of glass products is a slow and time-consuming process in the quality control process, so computer-aided systems, including image processing and machine learning techniques are used to overcome this problem. In this study, scratch and bubble defects of the jar, photographed in the studio with a white matte background and a-60 degrees peak angle, are investigated with the Yolo-V3 deep learning technique. Obtained performance is 94.65% for the raw data. Color space conversion (CSC) techniques, HSV and CIE-Lab Luv, are applied to the resulting images. V channels select for preprocessing. While the HSV method decreases the performance, an increase has been observed in the CIE-Lab Luv method. With the CIE-Lab Luv method, to which is applied the adaptive histogram equalization, the maximum recall, precision, and F1-score reach above 97%. Also, Yolo-V3 compared with the Faster R-CNN, it is observed that Yolo-V3 gave better results in all analyzes, and the highest overall accuracy is achieved in both methods when adaptive histogram equalization is applied to CIE-Lab Luv.eninfo:eu-repo/semantics/openAccessAdaptive Histogram EqualizationColor Space ConversionGlass Defect DetectionDeep LearningArticle39273173610.18280/ts.390238N/AWOS:000798489300038Q3