- 學位論文
智慧影像識別感測系統之進階應用研究
Advanced Application Research of Intelligent Image Recognition Sensing System
摘要
一般工業解決方案中，影像辨識訓練完成後能在產線達到一定的監控精確度，但根據操作環境的不同，解決方案所提供的變因可能尚有不夠完善之處，故本論文希望藉由使用近年崛起結合AI領域的智慧影像識別感測系統以驗證方法之可行性，並針對專門領域操作輔助與日常生活應用的方向進行延伸探討。 本論文中使用外部輔助紅光形成新的辨識特徵，研究智慧影像識別感測系統之進階應用。智慧影像識別感測系統採用基恩斯搭載AI 影像辨別感測器IV2系列。第一個部分探討琥珀色玻璃瓶內液位感測方案，藉由此感測系統的實驗量測驗證其效果，希望進一步用於提升影像識別感測領域在產線之應用；第二個部分探討光纖快接頭施作感測方案，藉由此感測器精細之分析能力，結合具專業領域知識人員，設計與規劃出本感測技術，提供即時提醒糾正使用者操作細節之系統。本方案可扮演輔助駕駛角色，藉由訓練快接頭施作流程偵錯特徵，領導新手學習及達成建立標準作業程序之功效。 在琥珀色玻璃瓶內液位感測方案中，已驗證所使用外部輔助紅光所產生之特徵具有幫助智慧影像識別感測系統辨識液位提昇辨識分數以及降低最高分數與最低分數差距之能力。在輔助紅光樣本訓練模型中，相較未加輔助光時更容易識別，進而提升辨識分數，在準確水位和照明模式長亮下，從最高分數99分及最低分數87分區間提升至最高100分與最低分數97分；在脈衝光模式下，從最高分數99分及最低分數87分區間提升至最高100分與最低分數95分；在無照明模式下，從最高分數91分及最低分數37分區間提升至最高98分與最低分數45分。 在光纖快接頭施作感測方案中，智慧影像識別感測系統具有專業領域知識，成功地判斷光纖快接頭內部光纖對接有無異常，並以OK與NG指示呈現。檢測出OK情況之光纖快接頭，其在1550nm波段的實測插入損耗約為0.4dB，符合光纖快接頭最大插入損耗0.5dB之規範。
並列摘要
In general industrial solutions, a certain level of monitoring accuracy can be achieved in the production line after finishing image recognition training. However, depending on the operating environment, the variables provided by the solution may need to be better. Therefore, this thesis hopes to use the intelligent image recognition sensing system that has emerged in recent years and combine it with the AI field to verify the method's feasibility and conduct extended discussions on the direction of operation assistance and daily life applications in specialized fields. This study uses external auxiliary red light to form new identification features and study the advanced application of intelligent image recognition sensing systems. The innovative image recognition sensing system uses KEYENCE-made AI image recognition sensor IV2 series. The first part discusses the liquid level sensing solution in amber glass bottles and verifies its effect through experimental measurements of this sensing system, hoping to improve further the application of image recognition sensing in production lines; the second part discusses the sensing solution for fiber optic fast connector implementation. By leveraging the sensor's precise analysis capabilities and combining it with professional domain knowledge, we design and plan this sensing technology to provide a system that instantly reminds users to correct operational details. This solution can coach and guide novices in learning and achieving the effect of establishing standard operating procedures by training quick joint operation process debugging features. In the liquid level sensing solution in the amber glass bottle, it has been verified that the characteristics generated by the external auxiliary red light can help the intelligent image recognition sensing system identify the liquid level, improve the recognition score, and reduce the gap between the highest score and the lowest score. In the auxiliary red light sample training model, it is easier to identify than without auxiliary light, thereby improving the identification score. Under the accurate water level and lighting mode, the maximum score of 99 points and the minimum score of 87 points increases to a maximum of 100 points and a minimum score of 97 points; in the pulse light mode, the maximum score of 99 points and the minimum score of 87 points increases to a maximum score of 100 points and a minimum score of 95 points; in the non-illumination mode, the maximum score of 91 points and the minimum score of 37 points increases to a maximum score of 98 points and a minimum score of 45 points. The intelligent image recognition system has professional knowledge of fiber optic fast connector installation sensing solutions. It successfully determines whether there are any abnormalities in the fiber optic docking inside the fiber optic fast connector and simultaneously displays an OK or NG state. The measured insertion loss of the fiber optic fast connector detected as OK in the 1550 nm band is about 0.4 dB, which meets the specification of the maximum insertion loss less than or equal to 0.5 dB.