人工智慧影像辨識在廠區安全偵測之應用案例研究

現今安全管理和災害預防的主要方式仍主要依靠管理人員的人工檢查，實務上還存在問題。即使有詳細的法律法規，降低現場風險並實施即時控制和預防仍然具有挑戰性。在工業4.0及智慧製造的趨勢下，人工智慧深度學習技術在電腦視覺、影像辨識上之快速發展，越來越多邊緣運算的發展，藉由 AI 晶片或邊緣運算裝置，使影像得以在邊緣端進行辨識。本研究論文將探討 AI 影像辨識在智慧工廠的應用的案例，包含瑕疵與品質檢測、物件偵測與辨識、員工安全行為、廠房安防監控，以及老舊設備儀表辨識等。藉由導入微型機器學習（TinyML）已成為一個新興領域，該技術的核心在於深度學習計算平台，這個平台不僅存在於雲端或伺服器上，還延伸到了人工智慧物聯網(AIoT)的感應器。這意味著以往需要透過雲端平台進行決策的任務，現在可以直接由 AIoT 感應器上的邊緣計算來完成，從而提供即時的影像識別結果。這不僅加快了決策的即時性，還有助於增強資訊的安全性。本研究針對相關應用討論所需架構、可能困難點與解決建議，以提升辨識的正確性與效率，可協助工廠安全管理人員改善廠區安全管理的不足之處。關鍵字: AI 影像辨識、智慧工廠、物件偵測辨識、案例研究、邊緣運算

關鍵字

AI 影像辨識；智慧工廠；物件偵測辨識；案例研究；邊緣運算

並列摘要

The primary means of safety management and disaster prevention in the contemporary era continue to rely heavily on manual inspections conducted by management personnel. This practice presents many problems. Despite the existence of comprehensive legislation and regulations, the reduction of on-site risks and the implementation of real-time controls and prevention remain challenging. The advent of Industry 4.0 and smart manufacturing has led to the rapid development of artificial intelligence and deep learning technologies in computer vision and image recognition, which has further advanced the emergence of edge computing. This has enabled image recognition to be performed at the edge using AI chips or edge computing devices. This research examines the AI image recognition in smart factories case study, including defect and quality inspection, object detection and recognition, employee safety behavior monitoring, factory security surveillance, and the identification of old equipment and instruments. The advent of TinyML has led to a new area of research, with the core of this technology lying in deep learning computational platforms. This platform is not limited to the cloud or servers; it also extends to sensors in the context of the Artificial Intelligence of Things (AIoT). This implies that tasks that previously necessitated decision-making via cloud platforms can now be accomplished directly by edge computing on AIoT sensors, resulting in real-time image recognition outcomes. This not only expedites the immediacy of decision-making but also enhances information security. This study examines the requisite architectural framework, potential obstacles, and proposed solutions for pertinent applications to enhance the precision and efficacy of recognition, thereby aiding factory safety management personnel in rectifying deficiencies in factory safety management. Keywords: AI Image Recognition, Smart Factory, Object Detection and Recognition, Case Study, Edge Computing