跳到主要內容

臺灣博碩士論文加值系統

(44.200.86.95) 您好!臺灣時間:2024/05/30 05:51
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:梁友銘
研究生(外文):LIANG, YU-MING
論文名稱:應用微型光譜感測晶片之光學非破壞式鮮乳新鮮度量測
論文名稱(外文):An Optical Non-destructive Milk Freshness Measurement Using Nano-spectrometer
指導教授:張正春張正春引用關係
指導教授(外文):CHANG, CHENG-CHUN
口試委員:高立人房同經邵皓強
口試委員(外文):KAO, LI-JENFANG, TUNG-CHINGSHAO, HAO-CHIANG
口試日期:2019-07-30
學位類別:碩士
校院名稱:國立臺北科技大學
系所名稱:電機工程系
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2019
畢業學年度:107
語文別:中文
論文頁數:69
中文關鍵詞:光譜法鮮乳新鮮度分類深度神經網路
外文關鍵詞:SpectroscopyMilk freshness classificationDeep neural networks
相關次數:
  • 被引用被引用:0
  • 點閱點閱:145
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
鮮乳擁有優良的動物性蛋白質且富含許多的營養,卻也是微生物容易孳生的溫床,不當的保存方式容易導致鮮乳腐敗變質。一般消費者雖可使用感官(眼、鼻)去評斷鮮乳是否新鮮,容易受個人主觀經驗影響,且在當消費者認為鮮乳發出酸臭味前,或許在鮮乳中的微生物菌數已經超標過量。有鑑於此,本研究希望透過光學式的新鮮度快速檢測方式判別出鮮乳是否已經不新鮮,以避免消費者誤食變質而不新鮮的鮮乳。
傳統量測鮮乳新鮮度大多是以鮮乳的微生物菌數和ph值作判別,當微生物菌數有明顯大量增生或ph值開始快速下降時,鮮乳通常已經放置接近十小時上下,一般民眾通常不會再進行飲用。而一般市售經超巴氏殺菌法(UHT-pasteurization)處理後的鮮乳,根據不同廠商規定,冷藏溫度須為7℃以下或是4 ℃以下,並標示離開冷藏後不能超過半小時,因此本研究目標透過使用光譜法,在鮮乳離開冷藏後,探討其在於攝氏約35±4℃每小時的光譜變化。
本論文處理光譜的方式可以分成3個階段,並在Python上分別使用Scikit-learn套件提供的機器學習演法和Tensorflow、Keras套件用深度學習的方式對鮮乳新鮮度進行判別。第一階段光譜預處理,我們將使用Z-score標準化對光譜訊號進行處理,減少不同樣本之間的光譜強度。第二階段光譜新鮮度分類和標籤,使用層次聚類(Hierarchical Clustering) 配合聚類評估演算法-輪廓係數,找出鮮乳光譜於時間上最適當的分群並給予標籤。第三階段使用KNN (k-Nearest Neighbors)、Random forest和DNNs (Deep neural networks)三種方式對置於攝氏約35±4℃下不同時間點的鮮乳做新鮮度分類預測。
由實驗結果顯示,使用KNN、Random Forest和DNN模型在廠牌A全脂鮮乳於6、24小時內和廠牌A低脂鮮乳於6小時內,都有很好的新鮮度預測準確率高達90%以上,且三者之間的預測正確率差異較為不大。而在廠牌A低脂鮮乳於24小時內鮮乳新鮮度預測中,當使用KNN和Random Forest於新鮮度預測準確率較不好時,分別為83.63%和82.06%,相對的使DNN突顯於鮮乳新鮮度預測的能力,總正確率約為87.67%。代表DNN於鮮乳新鮮度預測有一定的潛力。

Milk has excellent animal protein and is rich in many nutrients, but it is also a hotbed for microorganisms to be prone to birth. Improper preservation can easily lead to spoilage of milk. Although the consumer can use the senses (eyes, noses) to judge whether the milk is fresh, it is easily affected by the subjective experience of the individual, and before the consumer thinks that the milk gives a sour taste, perhaps the number of microbes in the milk has been exceeded. In view of this, this study hopes to determine whether the milk is not fresh through the optical rapid detection method to avoid consumers drinking the milk that is spoiled and not fresh.
The traditional freshness of milk is mostly judged by the microbial count and pH of milk. When the number of microbial bacteria is significantly increased or the pH value begins to decrease rapidly, the milk is usually placed close to ten hours, and people are usually not to drink. However, the milk that has been commercially processed by UHT-pasteurization is required to have a refrigerator temperature of 7 ° C or less or 4 ° C or less, and it should not exceed half an hour after leaving the refrigerator. Therefore, the aim of this study was to investigate the spectral changes over time at about 35 ± 4 ° C after milk was left refrigerated by using spectroscopy.
The way the thesis deals with the spectrum can be divided into three stages, and the freshness of the milk is discriminated in Python using the machine learning method provided by the Scikit-learn suite and the Tensorflow and Keras suites using deep learning. For the first stage of spectral preprocessing, we will use Z-score standardization to process the spectral signal to reduce the spectral intensity between different samples. The second stage of spectral freshness classification and labeling, using Hierarchical Clustering and clustering evaluation algorithm-contour coefficient, find out the most appropriate grouping of fresh milk spectrum in time and give labels. In the third stage, KNN (k-Nearest Neighbors), Random forest and DNNs (Deep neural networks) were used to predict the freshness of milk at different time points of about 35±4 °C.
The experimental results show that using KNN, Random Forest and DNN models in the brand A full fat fresh milk in 6 and 24 hours and the brand A low fat fresh milk in 6 hours, have a good freshness prediction accuracy more than 90%, and the difference in prediction accuracy between the three ways is relatively small.In the fresh milk freshness prediction of the brand A low-fat fresh milk in 24 hours, when using KNN and Random Forest, the accuracy of freshness prediction is not good, 83.63% and 82.06%, respectively. In contrast, DNN highlights the ability to predict fresh milk freshness with a total correct rate of approximately 87.67%.This means that DNN has a certain potential for fresh milk freshness prediction.

摘要 i
ABSTRACT iii
誌謝 v
目錄 vi
表目錄 ix
圖目錄 x
第一章 導論 1
1.1 研究動機與目的 1
1.2 研究內容介紹 2
1.3 論文架構 3
第二章 背景知識 4
2.1 鮮乳相關知識 4
2.1.1 鮮乳衛生標準 4
2.1.2 鮮乳傳統檢測方式 5
2.2 鮮乳新鮮度相關文獻探討 6
2.3 類神經網路應用於食物相關文獻 11
第三章 研究方法 13
3.1 量測平台製作 14
3.1.1 量測平台製作方法: 15
3.2 使用軟體介紹 20
3.3 光譜預處理 21
3.3.1 Z-score normalization 21
3.4 光譜分群後標籤 22
3.4.1 層次聚類(Hierarchical clustering) 22
3.4.2 聚類評估演算法-輪廓係數(Silhouette Coefficient) 23
3.5 新鮮度光譜分類 24
3.5.1 最近鄰居法(K-nearest Neighbors, K-NN) 24
3.5.2 隨機森林(Random Forest) 26
3.5.3 深度神經網路(Deep neural networks. DNNs) 28
第四章 實驗與結果 32
4.1 實驗環境參數設定 32
4.1.1 鮮乳樣本準備 32
4.1.2 量測平台光譜參數設定 33
4.2 鮮乳光譜量測實驗流程 34
4.3 鮮乳光譜樣本時間上分類選擇 35
4.4 實驗1:鮮乳光譜預處理 36
4.5 實驗2:鮮乳光譜新鮮度分類和標籤 37
4.5.1 廠牌A全脂鮮乳光譜6小時內分類 40
4.5.2 廠牌A全脂鮮乳光譜24小時內分類 41
4.5.3 廠牌A低脂鮮乳光譜6小時內分類 43
4.5.4 廠牌A低脂鮮乳光譜24小時內分類 44
4.6 實驗3:廠牌A全脂鮮乳6小時內分類預測結果 46
4.7 實驗4:廠牌A全脂鮮乳24小時內分類預測結果 50
4.8 實驗5:廠牌A低脂鮮乳6小時內分類預測結果 54
4.9 實驗6:廠牌A低脂鮮乳24小時內分類預測結果 58
第五章 結論與未來發展 62
5.1 結論 62
5.2 未來發展 64
參考文獻 65

[1]“鮮乳的營養成份.” [Online]
Available:http://www.dairy.org.tw/faq/?parent_id=30 [Accessed: May. 01, 2019]
[2]“過期牛奶中毒有哪些症狀.” [Online] Available: https://read01.com/zh-tw/DmmagQ.html#.W30y3OgzZPY [Accessed: Apr. 9, 2019]
[3]“全脂鮮奶喝了易胖?專家這樣解釋.” [Online]
Available:https://www.chinatimes.com/realtimenews/20190630001461-260405?chdtv [Accessed: July. 25, 2019]
[4]“鮮乳可以冷凍嗎,如何判斷鮮乳變質否?.” [Online]Available: http://www.poc-medical.url.tw/product_cg130310.html [Accessed: July. 25, 2019]
[5]“濃醇香就是好?鮮奶殺菌的祕密.” [Online]Available: https://www.pubu.com.tw/news/%E6%BF%83%E9%86%87%E9%A6%99%E5%B0%B1%E6%98%AF%E5%A5%BD-%E9%AE%AE%E5%A5%B6%E6%AE%BA%E8%8F%8C%E7%9A%84%E7%A5%95%E5%AF%86-4582 [Accessed: July. 25, 2019]
[6]“食品衛生標準:乳品類衛生標準.” [Online]Available: http://www.stst.com.tw/modules/news/article.php?storyid=26
[Accessed: July. 25, 2019]
[7]“全自動生乳總生菌數分析儀 BactoScan FC.” [Online]Available:  http://www.todays.com.tw/product-3a.asp?ser=817 [Accessed: July. 25, 2019]
[8]“飲料和食品酸度計/鹽度計 Acidity/Salinity Titrate for Drinks and Foods.” [Online]
Available:  http://www.greentree.com.tw/atf.htm [Accessed: July. 25, 2019]
[9]W. Ding, F. Gao and C. Yan, “LED-induced fluorescence spectroscopy technique for milk freshness detection” in Proc. of International Conference on Optical Communications and Networks (ICOCN), pp. 1-3, 2016.
[10]P. Silcock, M. Alothman, E. Zardin, S. Heenan, C. Siefarth, P. J. Bremer and J. Beauchamp, “Microbially induced changes in the volatile constituents of fresh chilled pasteurised milk during storage,” Journal of Food Packaging and Shelf Life, vol. 2, no. 2, pp. 81-90, 2014
[11]L. Liu, Z. Zhou, Z. Lv and D. Li, “Non-destructive Detection of Air Traces in the UHT Milk Packet by using Ultrasonic Waves,” Journal of Food Research International, vol. 38, pp. 631-638, 2005
[12]“破除網路謠言 各大廠牌鮮乳科學實測.” [Online]Available:  https://www.weichuan.com.tw/News/Detail/173 [Accessed: July. 25, 2019]
[13]D. D. Silalahi, C. E. Reaño, F. P. Lansigan, R. G. Panopio, N. C. Bantayan, “Using Genetic Algorithm Neural Network on Near Infrared Spectral Data for Ripeness Grading of Oil Palm (Elaeis guineensis Jacq.) Fresh Fruit,” Journal of Information Processing in Agriculture, vol. 3, pp. 252-261, 2016
[14]P. Wan, A. Toudeshki, H. Tan, R. Ehsani, “A methodology for fresh tomato maturity detection using computer vision,” Journal of Computers and Electronics in Agriculture, vol 146, pp. 43-50,2018
[15]J. Xinhua, X. Heru, Z. Lina, G. Xiaojing, W. Guodong, B. Jie, “Nondestructive detection of chilled mutton freshness based on multi-label information fusion and adaptive BP neural network,” Journal of Computers and Electronics in Agriculture, vol 155, pp. 371-377, 2018
[16]M. A. Putra, M. Rivai, A. Arifin, “Milk Assessment using Potentiometric and Gas Sensors in Conjunction With Neural Network,” in Proc. of 2018 International Seminar on Intelligent Technology and Its Applications (ISITIA), pp. 409-412, 2018
[17]D. Trientin, B. Hidayat, S, Darana, “Beef freshness classification by using color analysis, multi-wavelet transformation, and artificial neural network,” in Proc. of 2015 International Conference on Automation, Cognitive Science, Optics, Micro Electro-Mechanical System, and Information Technology (ICACOMIT), pp. 181-185, 2015
[18]“NSP32, Spectral Sensing Platform for IoT,” [Online]Available: https://nanolambda.myshopify.com/ [Accessed: July. 25, 2019].
[19]“[改善資料品質]Part-3 正規化與標準化資料_Z-score normalization.” [Online]Available:  https://ithelp.ithome.com.tw/articles/10202059 [Accessed: July. 11, 2019]
[20]“聚類——層次聚類Hierarchical Clustering.”[Online]Available:  https://www.itread01.com/content/1546543627.html[Accessed: July. 11, 2019]
[21]“Agglomerative hierarchical cluster tree.” [Online]Available:  https://ww2.mathworks.cn/help/stats/linkage.html#mw_08b425f7-fc8c-480a-b618-f768817e8e11[Accessed: July. 11, 2019]
[22]“聚類評估演算法-輪廓係數(Silhouette Coefficient).” [Online]Available:  https://www.itread01.com/content/1547114765.html [Accessed: July. 11, 2019]
[23]“幾分鐘了解一下K近鄰算法(KNN)原理及實踐.” [Online]Available:  https://kknews.cc/zh-tw/news/6bbonz3.html[Accessed: July. 20, 2019]
[24]“隨機森林原理.”[Online] Available: https://www.itread01.com/content/1542972489.html[Accessed: July. 20, 2019]
[25]“An end-to-end open source machine learning platform.” [Online]Available:  https://www.tensorflow.org/[Accessed: July. 20, 2019]
[26]“Keras: The Python Deep Learning library.” [Online]Available:  https://keras.io/[Accessed: July. 20, 2019]
[27]“乾貨|6 種激活函數核心知識點,請務必掌握!.” [Online]Available:  https://kknews.cc/zh-tw/other/ygkq3aa.html[Accessed: July. 20, 2019]
[28]“TensorFlow中的非饱和激活函数(Nonsaturating Activation Functions).”[Online]Available: https://www.twblogs.net/a/5b89ebb92b71775d1ce47c4f/zh-cn[Accessed: July. 20, 2019]
[29]“TensorFlow学习---tf.nn.dropout防止过拟合.”[Online] Available:  https://blog.csdn.net/huahuazhu/article/details/73649389[Accessed: July. 20, 2019]
[30]“[機器學習ML NOTE]SGD, Momentum, AdaGrad, Adam Optimizer.” [Online]Available:  https://medium.com/%E9%9B%9E%E9%9B%9E%E8%88%87%E5%85%94%E5%85%94%E7%9A%84%E5%B7%A5%E7%A8%8B%E4%B8%96%E7%95%8C/%E6%A9%9F%E5%99%A8%E5%AD%B8%E7%BF%92ml-note-sgd-momentum-adagrad-adam-optimizer-f20568c968db[Accessed: July. 20, 2019]
[31]“深度学习:原理简明教程09-深度学习:损失函数.” [Online]Available:  http://www.bdpt.net/cn/2017/12/24/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0%EF%BC%9A%E5%8E%9F%E7%90%86%E7%AE%80%E6%98%8E%E6%95%99%E7%A8%8B09-%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0%E6%8D%9F%E5%A4%B1%E5%87%BD%E6%95%B0/ [Accessed: July. 20, 2019]
[32]“深度学习技巧之Early Stopping(早停法).” [Online]Available:  https://blog.csdn.net/df19900725/article/details/82973049[Accessed: July. 20, 2019]
[33]“keras中的early stopping.” [Online]Available:  https://blog.csdn.net/ukakasu/article/details/80089866[Accessed: July. 20, 2019]

電子全文 電子全文(全文開放日期20240807,本篇電子全文限研究生所屬學校校內系統及IP範圍內開放)
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top