頁籤選單縮合
| 題 名 | 建立快速準確的臺灣與境外牛乳來源鑑定技術=Establishing Efficient and Accurate Techniques for Milk Origin Authentication |
|---|---|
| 作 者 | 凃柏安; 蕭振文; 葉亦馨; 林宗毅; 陳怡璇; 郭卿雲; 葉瑞涵; | 書刊名 | 中國畜牧學會會誌 |
| 卷 期 | 52:3 2023.09[民112.09] |
| 頁 次 | 頁191-207 |
| 分類號 | 439.6132 |
| 關鍵詞 | 鮮乳; 牛乳; 產地鑑別; 分類預測模型; Fresh milk; Milk; Origin authentication; Classification prediction model; |
| 語 文 | 中文(Chinese) |
| DOI | 10.30194/JCSAS.202309_52(3).0001 |
| 中文摘要 | 我國與紐西蘭於102年7月10日簽署「臺澎金馬個別關稅領域與紐西蘭經濟合作協定」(Agreement between New Zealand and the Separate Customs Territory of Taiwan, Penghu, Kinmen and Matsu on Economic Cooperation),紐西蘭液態乳品於2025年後將全面零關稅進口臺灣,國內市場將面臨國外乳品競爭的挑戰,因此為了保障消費者權益及國內酪農業,建立可靠的牛乳產地鑑別技術對國內酪農業的永續發展十分重要。本研究旨在通過分析乳成分和脂肪酸組成,建立一種快速準確的臺灣與境外牛乳產地鑑別方法。研究中收集了2022~2024年間市售的臺灣、日本、美國、澳洲及紐西蘭等國的牛乳樣品,並利用隨機森林機器學習模型進行產地預測。研究結果顯示,臺灣牛乳樣品在脂肪、乳糖和不飽和脂肪酸含量上顯著較境外牛乳高,而蛋白質、無脂固形物和尿素氮含量較低。通過隨機森林模型,訓練集和10次交叉驗證集的準確性分別達到100%和96.12%。模型在實際市場盲樣測試中達到了100%的正確分類,證明其在真實情境中具有高度準確性和可靠性,而具有鮮乳標章的30件國產鮮乳亦可全數驗證為臺灣牛乳,代表鮮乳標章為農政單位保護消費者權益所實施的有效行政管理措施。本研究之產地鑑別技術是建立於現行的分析技術及設備之上,因此能更可靠、更快速且低成本的與國內鮮乳產品銜接,有助於支持國內酪農業的永續發展。 |
| 英文摘要 | Taiwan signed the "Agreement between New Zealand and the Separate Customs Territory of Taiwan, Penghu, Kinmen and Matsu on Economic Cooperation" on July 10, 2013. Under this agreement, New Zealand liquid milk products will be imported to Taiwan tariff-free starting in 2025, posing significant competition to the domestic market. To protect consumer rights and the local dairy industry, it is crucial to establish reliable methods for authenticating the origin of milk, ensuring food safety, and promoting sustainable development in the dairy sector. We aims to develop a rapid and accurate method for identifying the geographical origin of milk by analyzing milk components and fatty acid composition. Samples of milk from Taiwan, Japan, the United States, Australia, and New Zealand were collected from 2022 to 2024 and analyzed using a random forest machine learning model for origin prediction. The results indicated that Taiwanese milk samples had significantly higher fat, lactose, and unsaturated fatty acid content, while protein, non-fat solids, and urea nitrogen content were lower. The model achieved accuracies of 100% in the training set and 96.12% in 10-fold cross-validation. In a real-world blind sample test, the model also achieved 100% correct classification, demonstrating high accuracy and reliability. Additionally, all 30 domestic fresh milk samples with the Fresh Milk Label were also verified to be of Taiwan origin, affirming the effectiveness of the administrative measures implemented by agricultural authorities to protect consumer rights. The low-cost testing technology and prediction model established in this study can be widely applied for batch testing of market products, effectively protecting consumer rights and supporting the sustainable development of the dairy industry in Taiwan. |
本系統中英文摘要資訊取自各篇刊載內容。