Oil Painting Defects Identification System with AI Techniques

This study uses Yolov5 object detection technology to detect the deterioration and defects of the painted layer of oil paintings. Two non-destructive detection light source image data, ultraviolet light and normal light will be used to identify refills, insect removal, and missing painted layers to assist Restoration engineers in speeding up the efficiency of missing annotation work in the work review process.This study applied the data augmentation method to increase the sample size and class balance. The results of this study show that the data augmentation method and class balancing can effectively improve the performance of the model; the performance of the proposed oil painting layer defect detection model has reached more than 98% in the class of insect exclusion and 70% in the class of retouching and painting layer loss. The study’s results can initially assist the restorer in the inspection process, effectively detect and locate defects, and provide the defect type and location to the restorer for reference, shortening the inspection process time and improving operational efficiency.