請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68511
標題: | 應用螢光光譜與螢光影像分析魚肉新鮮度之研究 Evaluation of Fish Freshness Using Fluorescence Spectroscopy and Fluorescence Imaging |
作者: | Li-Ming Chen 陳立銘 |
指導教授: | 林達德(Ta-Te Lin) |
關鍵字: | 激發發射矩陣,螢光影像,高光譜影像,魚肉新鮮度,手持式裝置, EEM,Fluorescence Imaging,Hyperspectral Imaging,Fish Freshness,Handheld Device, |
出版年 : | 2017 |
學位: | 碩士 |
摘要: | 近年來不論是國際或是台灣中,對於食品安全問題皆越來越重視,且政府對於食品內容物與食物中的生菌數與有害物質制定了相關規定。而現階段多為採取隨機抽樣檢查並進行耗時與複雜的化學分析實驗,因此主要針對大型公司與零售業者做抽樣檢查。為了使食安檢驗更為全面,行政院對於食品安全結合通訊科技推動了「食安五環」的計畫,將相關資料透過大數據分析,期望能夠主動預判可能的異常狀況。
為了實現食安五環中的「全民監督食安」與「十倍市場查驗十倍安全」的理念,本研究致力於建立一套輕便、便宜的手持式魚肉新鮮度檢測裝置。而為了建置此檢測裝置,本研究透過EEM獲得魚肉螢光指紋區域,在290 nm的激發光下可以獲得330~380 nm的發射光並在340~360 nm的激發光下可以獲得440~480 nm的發射光。本研究在氙氣燈前加裝340 nm帶通濾鏡,成功建立螢光高光譜影像,並將光譜資訊建立SVM模型。將台灣鯛魚新鮮度分為新鮮、室溫五小時、室溫十小時,可以獲得 77% 的準確率;而將海鱺分為新鮮、室溫五小時、室溫十小時,可以獲得 72% 的準確率。本研究欲比較可見光/近紅外光的高光譜影像分類準確度,因此也將其光譜資訊建立SVM模型,將新鮮度分為新鮮、室溫六小時、室溫十二小時的情況下,台灣鯛魚可以獲得 87% 的分類準確度、海鱺則是 65% 的分類準確度。透過螢光影像可以知道,在340 nm的激發光下,可以獲得470 nm波峰的發射光。因此在建立手持式系統時,便透過340 nm的LED激發光與搭配470 nm帶通濾鏡來實現其裝置。在室溫25℃下,台灣鯛魚的平均絕對誤差為2.2小時,而海鱺的平均絕對誤差為1.4小時。 Food safety issues are taken seriously in both Taiwan and international in recent year. The government develop the relevant provisions for food content, the number of bacteria and harmful substances in food. At this stage, random sampling, which is following by time-consuming and complex chemical analysis experiments are implemented mainly for large companies and retailers. In order to make the food safety inspection more comprehensive, the Executive Yuan combined food safety with communications technology and promote the 'Five rings of food security ' project. Expect to be able to take the initiative to predict possible abnormalities through the large data analysis of relevant information. This study is dedicated to the establishment of a lightweight, inexpensive hand-held fish freshness detection device in order to achieve the concept of “Citizen-monitoring-food-safety” and 'ten times the market check for ten times safety' in the food safety rings. To construct this detection device, we obtained the fish fluorescence fingerprint region through EEM, and obtained emission light of 330~380 nm at 290 nm excitation light and emission light of 440~480 nm at 340~360 nm excitation light. In this study, 340 nm bandpass filter was installed in front of the xenon lamp, the fluorescence hyperspectral image was successfully established, and use the spectral results to establish SVM model. The freshness is divided into fresh, room temperature for five hours and room temperature for ten hours, with 77% accuracy of Taiwan Tilapia, and 72% accuracy of Cobia . This study compare the hyperspectral image classification accuracy of visible / near-infrared light. Therefore, the SVM model is established under the spectral information. The freshness is divided into fresh, room temperature for six hours and room temperature for twelve hours. Taiwan Tilapia can get 87% classification accuracy, when cobia has 65% classification accuracy. Through the fluorescence imaging, you can get 470 nm peak emission light in the 340 nm excitation light. So the handheld device is composed of 340 nm LED excitation light with 470 nm bandpass filter in front of sensor. In room temperature, We can acquire the result of freshness models that the MAE value is 2.2 of Taiwan Tilapia and MAE value is 1.4 of cobia. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68511 |
DOI: | 10.6342/NTU201703925 |
全文授權: | 有償授權 |
顯示於系所單位: | 生物機電工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-106-1.pdf 目前未授權公開取用 | 5.34 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。