頁籤選單縮合
題 名 | AI影像辨識技術輔助博物館藏品盤點之初探--以國立臺灣歷史博物館為例=A Preliminary Study on AI Image Recognition Technology Assisting in Collection Inventory of Museum--A Case Study of the National Museum of Taiwan History |
---|---|
作 者 | 張鈞傑; 黃茂榮; 黃瀞慧; | 書刊名 | 科技博物 |
卷 期 | 28:2 2024.06[民113.06] |
頁 次 | 頁31-49 |
分類號 | 069.5 |
關鍵詞 | 文物物件偵測; AI影像辨識; 藏品盤點; 典藏管理; 智慧管理; Cultural relic object detection; AI image recognition; Collection inventory; Collection management; Smart management; |
語 文 | 中文(Chinese) |
中文摘要 | 本文主要討論國立臺灣歷史博物館(以下簡稱臺史博)為導入科技輔助藏品盤點,於2022年開發影像辨識盤點設備,講述建置此試驗設備之目的及實際運用,並介紹運用之技術與設計,再針對目前人力盤點作業流程進行分析,討論現階段影像辨識投入藏品盤點之可行性。依據公立博物館典藏品盤點作業辦法規定,公立博物館須對各自館內典藏品進行盤點,而依臺史博目前的館藏量每10年至少須將所有藏品盤點一次。截至2022年底,臺史博的藏品已逾14萬件,造成相當的管理壓力,因此嘗試以YOLOv5偵測演算法建置試驗型藏品辨識模型及設備,輔助藏品辨識與盤點,以期輔助內部管理及盤點工作推進。然而,辨識技術的發展雖日新月異,但是否真的已經達到期望中可減輕盤點人員負擔仍須再議,且臺史博的藏品中,有不少型制接近,如神像、粿餅印等,影像辨識是否能有效地的辨識其中的個體差異亦是未知數。因此,本文將臺史博所藏的一系列「粿餅印」選為測試對象,嘗試討論在以立體且個體間相似性較高的藏品作為盤點標的時,影像辨識技術如何輔助人力盤點程序更為流暢與精確,是否目前的影像辨識技術已臻成熟,進入可實際導入輔助盤點工作之階段。 |
英文摘要 | This article mainly discusses National Museum of Taiwan History (abbr. NMTH) introducing technology to assist collection inventory and develop image recognition inventory equipment in 2022. It describes the purpose and practical application of building this experimental equipment, and introduces the applied technology and design. Furthermore, it analyzes the current manual inventory process and discuss the feasibility of implementing image recognition for collection inventory at present. According to the Inventory Procedures for Collections in Public Museums, public museums must inventory their collections. NMTH has to inventory of all collections at least once every 10 years, in accordance with the current collection size. In the end of 2022, the collection of NMTH has exceeded 140,000 pieces, causing considerable management pressure. Therefore, it is trying to build experimental collection identification models and equipment using the YOLOv5 detection algorithm to assist in collection identification and inventory, expecting to assist internal management and promote inventory. However, although identification technology is developing at a rapid pace, it still needs to be discussed whether it has really reached the expectation of reducing the burden on inventory personnel. Moreover, there are many items in the collection of NMTH that are similar in shape, such as statues of gods, cake prints, etc. Whether images identification can effectively identify individual differences is also unknown. Therefore, this article selects a series of "cake prints" from the collection of NMTH as the test objects. It aims to discuss how image recognition technology can assist in making inventory procedures smoother and more accurate when dealing with three-dimensional collections that share high similarity. Additionally, it explores whether current image recognition technology has matured to the point where it can be practically integrated into inventory processes. |
本系統中英文摘要資訊取自各篇刊載內容。