<![CDATA[Pearls are highly valued gemstones, where quality is significantly determined by surface integrity, particularly the absence of hole defects. Traditional manual inspection is subjective and inefficient, creating a demand for automated, high-precision detection systems. This research develops an automated framework to identify hole-based defects in pearls sourced from the Bangsring Underwater region using digital image processing and machine learning. The methodology begins with preprocessing and thresholding to isolate pearl objects, followed by texture feature extraction using the Gray Level Co-Occurrence Matrix (GLCM). To handle high-dimensional data, this study utilizes Support Vector Machine - Recursive Feature Elimination (SVM-RFE) for feature selection, while classification is executed using an Artificial Neural Network (ANN). The experimental dataset consists of 240 high-resolution pearl images (120 with hole defects and 120 defect-free), evaluated across training cycles ranging from 100 to 1000. Results show that the stand-alone ANN achieved a peak accuracy of 95.83% at 1000 cycles, whereas integrating SVM-RFE enhanced performance to 97.62% at 900 and 1000 cycles. These findings confirm that the combined SVM-RFE and ANN framework provides a robust solution for automated defect detection, offering a scalable approach for industrial quality inspection in the Bangsring Underwater region.]]>
Copyrights © 2026
