結合物件偵測與深度學習應用於油畫彩繪層缺陷
|
| 摘要 機器學習、深度學習、物件偵測等相關技術應用已逐漸成熟，但是在文化資產保護與藝術品修復領域上，相關應用才正要開始。在作品修復狀況檢視流程中，過往都依賴修復師以人工方式標註缺失位置，導致有許多時間與精力耗費，且有時也會因為修復師的經驗而有誤判的可能。為解決上述問題，本研究提出以YOLOv5物件偵測技術，對油畫彩繪層進行缺陷檢測，輔助修復師加快作品檢視流程中缺失標註作業的效率，節省人工標註損害區域的時間，將更多的時間與心力投注在畫作修復作業；本研究使用紫外光與正常光二種非破壞性檢測光源，針對昆蟲排遺、彩繪層缺失進行辨識。研究結果表明，該方法可以輔助修復師標記與檢測油畫的缺陷，並提高了作業效率。希望藉此主題給予藝術修復領域一些技術輔助參考，提升修復的成效與減少人力時間的消耗。
關鍵字：YOLOv5、缺陷檢測、油畫彩繪層缺失、藝術品修復、非破壞性檢測。 ABSTRACT Machine learning, deep learning, object detection, and other related technologies have gradually matured, but their application in the field of cultural asset protection and art restoration is just beginning. In the past, the restoration process relied on the restorer to manually mark the missing locations, resulting in a lot of time and effort, and sometimes misjudgment due to the restorer's experience. To solve the above problems, this study proposes to use YOLOv5 object detection technology to detect defects in oil painting layers to assist restorers in speeding up the efficiency of defect marking during the work inspection process, saving time for manual marking of damaged areas, and devoting more time and effort to painting restoration operations. In this study, we used two non-destructive examination techniques. The results of this study show that this method can assist the restorer in marking and detecting defects in oil paintings and improve the efficiency of the work. It is hoped that this topic will provide some technical reference in the field of art restoration to improve the effectiveness of restoration and reduce the labor time consumption. KEYWORDS: YOLOv5; Defect Detection; Oil Painting Paint Layer Defect; Art Restoration; Non-Destructive Examination Techniques. |