酒瓶瑕疵及異物檢測之研究

Abstract

台灣菸酒股份有限公司因節能減炭政策，施行回收空瓶重覆使用。由於重覆使用關係，瓶子磨損度將會依回收次數逐漸嚴重，迫使包裝工廠使用的空瓶檢查機不得不調降剔除標準，反造成有問題之酒瓶漏逮。或因在製造過程中，所產生不同的問題，如破瓶、未封蓋等等。由於上述狀況，迫使酒廠須安排人力來做品管。有鑒於人員長時間接觸強光，易造成視力傷害，且包裝工廠現場環境，屬高分貝噪音場所，現場工作人員亦易造成聽力傷害。更常發生，因輸送帶高速運轉所造成人為疏忽漏逮，讓有瑕疵或瓶內有異物之瓶裝酒外流至消費市場，而造成客訴等負面形象。基於上述種種理由，促使本論文產生，其目的是要將現行人力品管朝自動化方向發展。 本論文主要在設計檢測回收清理過後之酒瓶瑕疵及異物辨識系統。將酒瓶置於測試平台上，以相機擷取測試平台上的影像，進而判斷出平台上的酒瓶是否為瑕疵品及瓶內是否有異物。會遭遇到的問題包括石英砂熔液於模具中成形，容易造成之酒瓶厚度不均、酒瓶的位移與旋轉、影像擷取時之光源不足，造成光線陰影的失真與雜訊等等問題。本論文主要是將原影像轉換成不同門檻值的二值影像，進而取得酒瓶中的異物及瑕疵特徵。最後，經由特徵比對，來辨識酒瓶是否為瑕疵品及瓶內是否有異物。經實驗證明本論文所使用的方法，有百分之98的準確率。

Taiwan tobacco and liquor corporation (TTL) has recycled and reused their bottles to support carbon reduction policies. Owing to the repeated use, these bottles get more and more abraded, which forces packaging factories to lower the exclusion criteria adopted by their empty bottle inspection (EBI) machines and thus some defective bottles pass this inspection. During the manufacturing process, some problems also occur, such as broken bottles and uncapped bottles. Therefore, factories have to allocate staff for quality control. The long-term strong light the staff is exposed to leads to eye damages while the noise at packaging factories also cause hearing damages to onsite workers. It’s even more frequent that the high-speed transmission of the conveyor belt sometimes makes workers fail in checking; the defective product or product with foreign matter inside get into consumer market, leading to negative images such as customer complaint. This thesis was inspired by reasons described above and aims to automatize the current manual quality control. This major purpose of this thesis is to design a system for detection of blemishes and foreign matter in recycled bottles that have been cleaned. These bottles are placed on a test platform with a camera used to capture bottle images, and these images will be converted into different threshold values binary images, which would be used to determine if the bottles on the platform are defective or if there is foreign matter inside. Finally, features are compared to determine whether a bottle contains foreign matter or is defective. The experimental results show that the proposed method reaches an accuracy rate of 98%