| 摘要: | 工業4.0時代的來臨，智慧製造的數位轉型趨勢已成為製造業提升競爭力的關鍵策略。在全球製造環境日益複雜且快速變化的背景下，標籤系統作為產品識別、追蹤與供應鏈管理的基礎，其效率與靈活性直接影響製造流程的整體表現。然而，傳統標籤設計與管理方法存在顯著缺陷：過度依賴人工編碼導致效率低下、專業軟體的高成本與有限授權造成資源瓶頸、標籤模板難以快速適應多樣化的客製需求，以及跨語系與國際標準的複雜要求增加了設計難度。這些問題不僅延長了產品上市時間，也增加了製造成本，成為製造業數位轉型過程中面臨的重大挑戰之一。傳統標籤設計依賴人工編碼與專業軟體，不僅耗時費力，每月人力成本高達10萬新台幣，且受限於軟體授權與專業人才的可用性。因此，本研究基於以上製造業產業現況所面臨的挑戰，提出一套整合YOLO深度學習模型與圖片轉程式碼技術的智慧製造標籤系統。本研究首先系統性比較YOLOv7、YOLOv8與YOLOv9三種模型在標籤辨識任務的表現，找出最適合的影像辨識模型；其次，開發圖片轉程式碼技術，實現從標籤視覺元素到程式碼的自動轉換；最後，整合兩項技術為標籤自動化系統。研究結果顯示，表現最佳的YOLO v9模型在標籤元素辨識上達到88.9%的準確率，而本研究提出之智慧製造標籤系統將標籤設計時間從平均2-8小時縮減至10分鐘，有效降低了約97%的設計成本。本研究智慧製造標籤系統的開發有效以資料驅動為基礎的深度學習技術實現數位轉型，透過程式碼自動化生成的流程最佳化以及透過數位化技術降低對實體資源依賴的虛實整合。透過智慧製造標籤系統的實際應用，不僅為製造業解決了傳統方法的效率與成本問題，亦可謂組織本身帶來多重的策略優勢，如：實現標籤設計的民主化，使非專業人員也能創造高品質標籤；提高製造靈活性，使企業更快速響應市場變化；最佳化資源配置，讓工程師更專注於具專業性與創新性的任務；提升標籤一致性與準確性，減少因人為錯誤導致的品質問題；同時促進製造資料的標準化與整合，為後續的大數據分析與智慧決策奠定基礎。本研究不僅為製造業標籤管理提供了實用的自動化解決方案，也為圖片轉程式碼技術在製造業的應用提供了個案實證，更為製造業提供明確的數位轉型可複製技術路徑指引。;The emergence of Industry 4.0 has positioned digital transformation in smart manufacturing as a vital strategy for manufacturing competitiveness. In today′s complex and dynamic global manufacturing landscape, label systems underpin product identification, tracking, and supply chain management, with their efficiency directly affecting overall manufacturing performance. Traditional label design and management methods face several limitations: dependency on manual coding reduces efficiency, expensive specialized software with limited licenses creates bottlenecks, templates fail to adapt to diverse customization needs, and multilingual requirements complicate design processes. These issues extend time-to-market, increase costs, and present significant barriers to manufacturing digital transformation. Current approaches rely heavily on manual processes that cost approximately NT$100,000 monthly and are constrained by both software availability and specialized skill requirements. This study addresses these industrial challenges by developing a smart manufacturing label system that integrates YOLO deep learning models with image-to-code technology. The study compares YOLO v7, YOLO v8, and YOLO v9 performance in label recognition tasks, develops technology for converting visual elements to program code, and combines these technologies into an automated system. Findings indicate that the most effective YOLO variant achieved 88.9% accuracy in element recognition, while the integrated system reduced design time from 2-8 hours to 10 minutes, decreasing design costs by approximately 97%. This system implementation advances digital transformation through data-driven deep learning techniques, optimizes processes through automated code generation, and reduces dependency on physical resources through digital integration. The practical applications provide numerous strategic advantages: enabling non-specialists to create high-quality labels, increasing manufacturing responsiveness, redirecting engineering talent to more innovative tasks, improving label consistency, and standardizing manufacturing data to support advanced analytics. This research offers both a practical solution for label management and empirical evidence for image-to-code applications in manufacturing, ultimately providing a reproducible pathway for digital transformation in manufacturing operations. |
