請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/59961
標題: | 線掃描式高光譜影像系統之開發-以葉菜硝酸鹽分布分析之應用為例 Development of a Line-scan Hyperspectral Imaging System - An Example for Distribution Analysis of Nitrate Content in Leafy Vegetables |
作者: | Sheng-Yung Lin 林聖詠 |
指導教授: | 陳世銘(Su-Ming Chen) |
關鍵字: | 近紅外光,高光譜影像系統,葉菜,硝酸鹽,成分可視化, Near-infrared,Hyperspectral Imaging System,Leafy vegetable,Nitrate,Content visualization, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 硝酸鹽為葉菜生長必須之氮肥的主要來源,生長過程中過量施肥使作物產量提升,也增加了植物體內的硝酸鹽濃度,人體若食用過量的硝酸鹽導致血液疾病和罹癌。目前標準的硝酸鹽檢測方法雖然精準但耗時長,故僅能以抽樣的方式來進行檢測。快速準確且非破壞性的檢測對於農作物等農產品及生物材料之檢測非常重要,而光譜檢測技術符合了這些要求。生物材料變異性大,能結合光譜與空間資訊的高光譜影像技術逐漸被廣泛應用。相較於其他拍攝方法,線掃描的方式擁有低雜訊干擾以及高影像掃描速度,因此特別適合即時產品檢測。 為了能快速且精準的拍攝農產品高光譜影像,本研究整合了線高光譜相機、滾珠螺桿滑台、步進馬達、鹵素光源、控制器等硬體元件,製作了一可感測950 ~ 1700 nm波段之線掃描式高光譜影像系統並使用LabVIEW與MATLAB進行系統軟體建置。相機幀數最高可達344 FPS,拍攝一次範圍300 mm × 260 mm之空間的高光譜影像只需花費15秒。系統相當穩定,系統光雜訊之平均雜訊偏移(Bias)小於2.36 × 10^-3 Abs、均方根值(RMS)小於8.49 × 10^-5 Abs。而系統軟體功能包含光場調整人機介面、自動化送樣速度校正、高光譜影像自動化流程拍攝、光譜資訊互動式讀取人機介面與成分可視化影像顯示等功能,可提供光譜分析研究快速且自動化光譜資料取得。 本研究選擇硝酸鹽含量較高、十字花科葉菜類的青江菜做為實驗對象,以光譜吸收度來預測葉片之硝酸鹽濃度,模型建立分別使用MLR與MPLSR來進行迴歸。最好之預測結果以8次重複取樣之光譜使用MLR方法建立之模型,校正組之rc2可達0.85、SEC為684 mg/kg,而交叉驗證組之rcv2可達0.83、SECV為713 mg/kg。此結果也證明增加重複取樣次數來降低光雜訊之影響可以提高預測結果。從青江菜硝酸鹽濃度可視化影像中可以確定葉菜各組織與位置的濃度分布並不均勻,濃度由高到低分別為葉脈 > 中心葉肉 > 外緣葉肉。此外,本研究成功使用可視化影像進行後續試驗,確認在葉菜採收後置於冷藏狀態下,仍可以透過4小時的短期光處理使硝酸鹽快速下降約20%。 本研究線掃描式高光譜影像系統與儀器級分光光度計在光譜分析青江菜硝酸鹽濃度上之預測能力相當,並且能將高光譜影像經檢量線轉換後顯示成分可視化影像。可視化影像包含空間與成分之資訊,非常適合應用於需觀察成分動態變化之研究上。未來,此系統將繼續用於建立光譜成份檢量線以及農產品成份影像可視化之研究。 Nitrate, as the main source of nitrogen, is essential for plant growth. Excessive application of nitrogen fertilizer could increase plant yield, but also lead to elevated nitrate concentration in leafy vegetables. Consuming high doses of nitrate may be associated with blood diseases and cancer. The current standard approaches used for nitrate detection, although accurate, are destructive, time-consuming and limited to the small size of samples. The characteristics of fast, accurate and non-destructive are the important requirement for the quality assessment of agricultural products and biomaterials, and the spectral detection technology meets these requirements. Hyperspectral imaging technology, combining both the spatial and spectral information of a sample, has been widely applied in biomaterial detection. Compared to other scanning methods, the line-scan method has the advantages of low noise and high scanning speed, particularly suited for real-time detection. To acquire the hyperspectral images of the agricultural products in a fast and accurate manner, this study developed a hyperspectral imaging system (HIS) using a hyperspectral camera with 950 - 1700 nm detection wavelength, and integrated with hardware of stepper motor, ball screw slide, halogen light source and controller. A control program was written by LabVIEW and MATLAB software. The system’s frame rate can reach up to 344 frames per second, with only 15 seconds are required to scan an area of 300 mm × 260 mm. The stability of the system is determined by measuring the photometric noise. The bias of noise is less than 2.36 × 10^-3 Abs and the root mean square of noise is less than 8.49 × 10^-5 Abs. The functionality of the developed system software including the light field adjustment, automatic correction for sample delivery speed, automatic hyperspectral image acquisition, interactive interface of spectral information and visual image display of ingredients which can provide fast and automated hyperspectral data acquisition for spectral analysis research. In this study, bok-choy (Brassica chinensis Linn), commonly with high nitrate content, was used as the study material. The nitrate concentration in bok-choy leaves was predicted by absorption spectrum. The models, specifically, multiple linear regression (MLR) and modify partial least squares regression (MPLSR), were used to establish the prediction model. The best prediction model was given by MLR with 8 repetitions of sample scanning. The rc2 and SEC of calibration samples were 0.85 and 684 mg/kg, respectively, while the rcv2 and SECV of the cross-validation samples were 0.83 and 713 mg/kg, respectively. This result also indicated that increasing the number of sample scanning to reduce the influence of optical noise can improve the prediction result. From the visualization of nitrate concentration in bok-choy, it can be observed that the nitrate concentration was not uniformly distributed in the leaf tissues, from high to low: leaf veins > center of leaf > edge of the leaf. Also, this study showed that the nitrate concentration of post-harvested leafy vegetables can be significantly reduced by about 20% through the short-term light treatment of 4 hours under cold storage. The developed line-scan HIS in this study is comparable to the instrument-level spectrophotometer for the spectral analysis of nitrate concentration in bok-choy. The nitrate content obtained by the hyperspectral image can be visualized as a distribution map using the calibration line. The visualized image contains information about space and content concentration, which is very suitable for research that needs to observe the dynamic changes of concentration. For the future, this system will continually research on establishing the spectral prediction models and visualization of the spectral images. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/59961 |
DOI: | 10.6342/NTU202003268 |
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 | |
---|---|---|---|
U0001-1308202015221000.pdf 目前未授權公開取用 | 7.04 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。