基于四方對(duì)稱光源透射光譜的臍橙可溶性固形物檢測(cè)

宋 杰，李光林，楊曉東，張 信，劉旭文

基于四方對(duì)稱光源透射光譜的臍橙可溶性固形物檢測(cè)

宋 杰，李光林※，楊曉東，張 信，劉旭文

（西南大學(xué)工程技術(shù)學(xué)院，重慶 400715）

提高利用可見-近紅外（Vis-NIR）透射光譜檢測(cè)臍橙內(nèi)部物質(zhì)含量的準(zhǔn)確性在生產(chǎn)實(shí)際中具有重要意義。該研究利用特制的可見-近紅外透射光譜測(cè)量裝置采集了199個(gè)福本臍橙果蒂向上、水平、向下3種位置的透射光譜，比較了多元散射校正（multivariate scattering correction, MSC）、標(biāo)準(zhǔn)正態(tài)變量變換（standard normal variate transformation, SNV）、一階導(dǎo)數(shù)和二階導(dǎo)數(shù)預(yù)處理的效果，并采用效果最好的一階導(dǎo)數(shù)對(duì)透射光譜進(jìn)行預(yù)處理。在此基礎(chǔ)上，結(jié)合后向區(qū)間偏最小二乘法（backward interval partial least squares, BiPLS）優(yōu)選特征波段，競(jìng)爭性自適應(yīng)重加權(quán)采樣（competitive adaptive re-weighted sampling, CARS）挑選特征變量建立了基于果蒂向上、水平、向下3種位置各自的透射光譜以及3種位置的平均光譜和加權(quán)光譜的可溶性固形物（soluble solid content, SSC）的偏最小二乘（partial least squares, PLS）模型。在果蒂向上、水平、向下3種位置各自的透射光譜建立的PLS模型中，基于果蒂水平位置透射光譜的PLS模型最優(yōu)，校正相關(guān)系數(shù)為0.914，校正均方根誤差為0.380，預(yù)測(cè)相關(guān)系數(shù)為0.924，預(yù)測(cè)均方根誤差為0.404。基于果蒂向上、水平、向下3種位置平均透射光譜和加權(quán)透射光譜建立的PLS模型均取得了較好的預(yù)測(cè)結(jié)果，預(yù)測(cè)相關(guān)系數(shù)均大于0.91，預(yù)測(cè)均方根誤差均小于0.43。該研究可以為臍橙內(nèi)部物質(zhì)含量在線檢測(cè)裝備的研制提供參考。

果實(shí)；光譜分析；模型；臍橙；四方對(duì)稱光源；透射；可溶性固形物

0 引 言

臍橙是一種營養(yǎng)豐富、經(jīng)濟(jì)價(jià)值較高的水果，在中國有廣泛的種植面積。隨著生活水平的提高，人們?cè)谫徺I水果時(shí)不僅僅局限于對(duì)外觀品質(zhì)的要求，而是更加關(guān)注其內(nèi)部品質(zhì)[1-2]。因此渠道商越來越注重臍橙的產(chǎn)后分級(jí)處理，提高臍橙的商品化價(jià)值。

可見-近紅外光譜（Vis-NIR）技術(shù)是一種快速、非破壞性檢測(cè)技術(shù)，已廣泛應(yīng)用于水果內(nèi)部品質(zhì)的無損檢 測(cè)[3-5]，如蘋果[6-9]、桃子[10-11]、西瓜[12-13]、梨[14-15]等。可溶性固形物（soluble solid content, SSC）是水果內(nèi)部品質(zhì)評(píng)價(jià)的主要指標(biāo)之一。在利用Vis-NIR技術(shù)實(shí)現(xiàn)臍橙SSC檢測(cè)的研究中，Jamshidi等[16]和Ncama等[17]采用反射測(cè)量方式沿臍橙赤道部位采集多點(diǎn)光譜求平均建立PLS模型對(duì)臍橙SSC進(jìn)行檢測(cè)。由于反射測(cè)量中光的穿透深度有限，而可溶性固形物在臍橙中沿橫徑和縱徑分布都不均勻[18]，因此利用反射測(cè)量方式獲取的光譜信息不能承載臍橙果實(shí)內(nèi)部的SSC信息。

近年來部分研究人員利用透射測(cè)量方式獲取臍橙果實(shí)內(nèi)部的光譜信息進(jìn)行分析。劉燕德等[19-20]利用水果分級(jí)傳送帶，通過人工上果將臍橙放置在果托上，果蒂與果臍連線方向與傳送帶運(yùn)動(dòng)方向一致，采集赤道部位的光譜對(duì)臍橙SSC進(jìn)行動(dòng)態(tài)檢測(cè)。由于果皮對(duì)光的衰減作用較強(qiáng)[21-22]，并且果皮在果臍、赤道、果蒂部位的厚度不均勻，導(dǎo)致利用透射測(cè)量方式獲取臍橙內(nèi)部物質(zhì)含量的光譜信息時(shí)會(huì)受到光源布置和臍橙放置位置的影響。許文麗等[23]使用了2個(gè)鹵鎢燈水平布置于樣品兩側(cè)，實(shí)現(xiàn)半透射檢測(cè)，研究了果蒂與果臍連線與入射光線平行、垂直和任意角度時(shí)對(duì)近紅外光譜檢測(cè)結(jié)果的影響。然而，由于臍橙形狀大小不一致，并且果皮厚度不均勻，僅通過兩側(cè)光源進(jìn)行輻射時(shí)還存在一定的測(cè)量誤差。為了獲取臍橙內(nèi)部更全面的透射光譜信息，進(jìn)一步降低測(cè)量誤差，本文通過自己設(shè)計(jì)的光源系統(tǒng)將4個(gè)鹵鎢燈對(duì)稱布置于樣品四周，光源中心軸線與信號(hào)探頭豎直中心軸線所成角度為40°，且光源中心軸線的交點(diǎn)為樣品下部，通過樣品下部接收透射信號(hào)。比較研究了基于果蒂向上、水平、向下3種位置透射光譜的PLS模型對(duì)臍橙SSC進(jìn)行預(yù)測(cè)的效果。在此基礎(chǔ)上，研究了利用3種位置平均光譜和加權(quán)光譜建模對(duì)臍橙SSC進(jìn)行預(yù)測(cè)的效果，以期為臍橙內(nèi)部物質(zhì)含量在線檢測(cè)裝備的研制奠定更多的基礎(chǔ)。

1 材料與方法

1.1 臍橙樣品

于2017年12月20日在重慶巫山（經(jīng)度：109.86，緯度：31.10）采摘了199個(gè)福本臍橙樣品進(jìn)行試驗(yàn)。樣品運(yùn)輸至實(shí)驗(yàn)室之后，剪掉枝葉，保留果蒂，每個(gè)臍橙均編號(hào)并裝入厚度為0.02 mm的聚乙烯袋，保存于溫度為4 ℃，相對(duì)濕度為60%的儲(chǔ)藏室中，保存時(shí)間不超過 3 d。在進(jìn)行試驗(yàn)前將樣品放置于室溫環(huán)境（14±2）℃ 24 h。

1.2 臍橙透射光譜測(cè)量

臍橙的透射光譜是通過我們自己設(shè)計(jì)的光譜測(cè)量系統(tǒng)采集的，測(cè)量系統(tǒng)包括一個(gè)580 mm × 600 mm × 610 mm的箱體用于隔離系統(tǒng)外部的光，箱體固定，并在底部設(shè)置有防震墊。箱體內(nèi)部設(shè)置有隔離板，防止光源對(duì)信號(hào)接收探頭造成干擾，隔離板中間設(shè)置有一個(gè)帶通孔的果托（孔徑為44 mm）。光源包括4個(gè)24 V/100 W的鹵鎢燈泡（OSRAM, 64460U, 3 000 K, Germany），對(duì)稱設(shè)置在果托上方。通過試驗(yàn)測(cè)試，確定光源中心軸線與信號(hào)探頭豎直中心軸線所成角度為40°，燈泡中心與待測(cè)臍橙中心的水平距離為100 mm，垂直距離為80 mm時(shí)，獲取的透射信號(hào)最強(qiáng)。為了達(dá)到更好的聚光效果，使用了發(fā)散角為10°的燈杯。為了便于散熱，燈頭處連接了一個(gè)直徑為110 mm，厚度為10 mm的環(huán)形散熱器（60齒結(jié)構(gòu)），連接處填充了導(dǎo)熱硅脂。2個(gè)風(fēng)扇（Sanyo GV1224P1H03, Japan）正對(duì)設(shè)置在箱體左右兩側(cè)，與散熱器位置匹配，一個(gè)用于吸入箱體外的冷空氣，另一個(gè)用于排出箱體內(nèi)的熱空氣。電源（WEHO, SCN-800-24, China）、光譜儀（Ocean Optics, QEpro, USA，響應(yīng)范圍400~980 nm）、信號(hào)接收探頭（視場(chǎng)角約43°）設(shè)置在隔離板下面，與光源完全隔離。使用了一臺(tái)安裝有光譜采集軟件（OceanView1.6.3）的計(jì)算機(jī)（Lenovo, IdeaPad 710S, China）采集光譜數(shù)據(jù)。臍橙透射光譜測(cè)量系統(tǒng)原理圖如圖1所示。

光譜儀的積分時(shí)間設(shè)置為50 ms，在采集透射光譜前保存了參考光譜和暗光譜，則透過率可表示為

根據(jù)下式將透過率轉(zhuǎn)化為吸光度：

式中為吸光度。

分別采集了臍橙3個(gè)位置的透射光譜，包括果蒂向上（P1）、果蒂水平（P2）、果蒂向下（P3）。在每個(gè)位置沿水平方向旋轉(zhuǎn)臍橙，每間隔120°采集一次光譜，共采集3次取平均，如圖2所示。

圖2中可以看到幾個(gè)明顯的特征吸收峰，出現(xiàn)在680 nm附近的峰是由于色素吸收所致[24]，760 nm附近的吸收峰是OH鍵伸縮振動(dòng)和水的吸收的四倍頻[25]，970 nm附近的吸收峰是OH鍵伸縮振動(dòng)和水的吸收的三倍頻[26]。圖2c中存在幾條光譜與正常光譜發(fā)生偏離，這幾個(gè)樣品屬于畸形果，果形不正，為了防止漏光，在擺放該臍橙測(cè)量果蒂向下位置的透射光譜時(shí)，果蒂并不是朝向正下方，而是有些偏離，導(dǎo)致了這幾個(gè)樣品在該位置的透過率偏高，從而導(dǎo)致光譜曲線發(fā)生偏離。考慮到實(shí)際應(yīng)用中存在畸形果對(duì)測(cè)量的影響，為了保證模型的適應(yīng)能力，我們?cè)诮r(shí)并未將這些偏離的光譜數(shù)據(jù)進(jìn)行剔除。

1. 風(fēng)扇 2. 光源 3. 臍橙 4. 散熱器 5. 隔板 6. 果托 7. 探頭 8. 光譜儀 9. 計(jì)算機(jī)

1.3 樣品理化參數(shù)測(cè)量

采集完光譜后，測(cè)量了樣品的理化參數(shù)。通過游標(biāo)卡尺測(cè)量了樣品的橫徑和縱徑，以及樣品在果臍部位、赤道部位和果蒂部位的皮厚。之后用榨汁機(jī)（KESUN, KP60SC, China）將樣品和皮一同榨成汁。將汁液過濾后利用數(shù)字折射計(jì)（ATAGO, RX-5000i-Plus, Japan）測(cè)量了樣品的SSC含量。理化參數(shù)統(tǒng)計(jì)結(jié)果如表1所示。

表1中臍橙樣品的橫徑平均值和縱徑平均值相差不大，表明該品種臍橙果形為類球形。從臍橙不同部位皮厚統(tǒng)計(jì)結(jié)果可以看出，果蒂部位的皮最厚。

1.4 算法原理

1.4.1 后向區(qū)間偏最小二乘法

1.4.2 競(jìng)爭性自適應(yīng)重加權(quán)算法

競(jìng)爭性自適應(yīng)重加權(quán)算法（competitive adaptive re-weighted sampling, CARS）的采樣過程類似于達(dá)爾文進(jìn)化論中的“適者生存”原則，它是以一種有效和競(jìng)爭的方式實(shí)現(xiàn)的[29]。假設(shè)蒙特卡羅采樣次數(shù)為，然后CARS依次從次蒙特卡羅采樣中以迭代和競(jìng)爭的方式選擇個(gè)波長子集。在CARS中，PLS模型的回歸系數(shù)的絕對(duì)值被用來作為評(píng)估每個(gè)變量重要性的指標(biāo)。在每次采樣中，以固定比例隨機(jī)選擇樣本建立校正模型，然后采用指數(shù)遞減函數(shù)和自適應(yīng)重加權(quán)算法根據(jù)回歸系數(shù)選擇關(guān)鍵波長，最后通過PLS交叉驗(yàn)證來評(píng)估所選擇的數(shù)據(jù)集[30]。

注：圖2a、2b、2c中從左至右依次為光譜采集位置俯視圖、光譜采集位置正視圖、該位置的平均透射光譜，圖中A、B、C為臍橙每旋轉(zhuǎn)120°所對(duì)應(yīng)的位置。

表1 福本臍橙理化參數(shù)統(tǒng)計(jì)結(jié)果

2 結(jié)果與分析

利用軟件The Unscrambler X 10.4，采用outlier方法剔除了一個(gè)異常樣品，并將剩余樣品隨機(jī)劃分為校正集（150個(gè)）和預(yù)測(cè)集（48個(gè)）。分別采用SNV、MSC、一階導(dǎo)數(shù)和二階導(dǎo)數(shù)對(duì)臍橙透射光譜進(jìn)行預(yù)處理，并通過PLS交叉驗(yàn)證來評(píng)估預(yù)處理效果，如表2所示。

表2中，P1、P2、P3的加權(quán)系數(shù)是根據(jù)P1、P2、P3光譜的交叉驗(yàn)證結(jié)果，采用貢獻(xiàn)率的思路來確定的。P2光譜的交叉驗(yàn)證結(jié)果明顯優(yōu)于P1和P3，則進(jìn)行加權(quán)時(shí)權(quán)重應(yīng)最大，P3光譜的交叉驗(yàn)證結(jié)果最差，則進(jìn)行加權(quán)時(shí)權(quán)重應(yīng)最小。分別以0.05為取值間隔進(jìn)行試驗(yàn)，并比較PLS交叉驗(yàn)證結(jié)果，最終確定最佳系數(shù)為0.15、0.8和0.05。從表2可以看出，采用SNV、MSC、一階導(dǎo)數(shù)和二階導(dǎo)數(shù)對(duì)P1、P2、P3位置的光譜及三者的平均光譜和加權(quán)光譜進(jìn)行預(yù)處理時(shí)，均是一階導(dǎo)數(shù)效果最佳，但由于不同位置采集的透射光譜的吸光度和信噪比有差異，因此進(jìn)行預(yù)處理時(shí)通過調(diào)整窗口數(shù)來獲得最佳預(yù)處理結(jié)果。4種預(yù)處理方法中，采用SNV和MSC預(yù)處理的結(jié)果相差不大，這與文獻(xiàn)報(bào)道的結(jié)論一致[31]。采用二階導(dǎo)數(shù)預(yù)處理后PLS模型的因子數(shù)少于一階導(dǎo)數(shù)預(yù)處理后的PLS模型，但是交叉驗(yàn)證結(jié)果比一階導(dǎo)數(shù)預(yù)處理略差。P1、P2、P3位置的光譜及三者的平均光譜和加權(quán)光譜經(jīng)一階導(dǎo)數(shù)預(yù)處理后建立的PLS模型中，基于P1、P2和P3位置加權(quán)光譜的PLS模型最優(yōu)。將上述經(jīng)一階導(dǎo)數(shù)預(yù)處理后的光譜數(shù)據(jù)用于后續(xù)分析，為便于表示，記(P1+P2+P3)/3為P4，（0.15P1+0.8P2+0.05P3）為P5，采用BiPLS優(yōu)選特征波段，將P1、P2、P3、P4、P5的光譜數(shù)據(jù)分別分為30～50段（間隔為5段），并基于PLS模型的RMSECV評(píng)估分段效果。當(dāng)分段數(shù)分別為35、40、30、35、40時(shí)效果最佳，統(tǒng)計(jì)結(jié)果如表3所示。

表2 基于不同預(yù)處理方法的PLS交叉驗(yàn)證結(jié)果

注：表2中P1、P2、P3分別代表臍橙果蒂向上、果蒂水平、果蒂向下位置獲取的透射光譜，(P1+P2+P3)/3代表P1、P2、P3的平均光譜，0.15P1+0.8P2+0.05P3代表P1、P2、P3的加權(quán)光譜。

Note: P1, P2 and P3 represent the transmission spectra of navel orange pedicle upward, pedicle horizontal placed and pedicle downward respectively; (P1+P2+P3)/3 represents the average spectra of P1, P2 and P3; 0.15P1+0.8P2+0.05P3 represents the weighted spectra of P1, P2 and P3.

表3 基于BiPLS的特征波段優(yōu)選結(jié)果

從表3中可以看出，基于P5光譜的波段優(yōu)選效果最佳，保留的變量數(shù)為170，RMSECV為0.390。基于P2光譜的波段優(yōu)選效果與基于P5光譜的結(jié)果很接近，保留的變量數(shù)為180，RMSECV為0.393?；赑4光譜的波段優(yōu)選效果次之，RMSECV為0.458，但保留的變量數(shù)為308，數(shù)量較多?；赑3光譜的波段優(yōu)選效果最差，保留的變量數(shù)為114，RMSECV為0.837，這可能是由于果蒂部位的皮較厚（見表1），且果蒂的存在對(duì)光的穿透形成了較大障礙，使得透射光譜的信噪比較低所致。

在對(duì)P1、P2、P3、P4、P5光譜優(yōu)選波段的基礎(chǔ)上，分別利用CARS算法挑選特征變量并建立PLS模型，其中CARS運(yùn)行次數(shù)均為20次，統(tǒng)計(jì)最優(yōu)結(jié)果如表4所示：

表4中，基于P5光譜建立的PLS模型最優(yōu)，校正相關(guān)系數(shù)為0.914，校正均方根誤差為0.382，預(yù)測(cè)相關(guān)系數(shù)為0.928，預(yù)測(cè)均方根誤差為0.383。但該模型是通過果蒂向上、水平、向下3個(gè)位置加權(quán)透射光譜建立的，生產(chǎn)中需利用特殊裝置進(jìn)行定位，在線檢測(cè)較難實(shí)現(xiàn)。基于P4光譜建立的PLS模型也取得了較好的預(yù)測(cè)結(jié)果，預(yù)測(cè)相關(guān)系數(shù)為0.911，預(yù)測(cè)均方根誤差為0.420。該模型只利用了果蒂向上、果蒂水平和果蒂向下3個(gè)位置光譜的平均光譜建模，而實(shí)際應(yīng)用時(shí)可以通過翻滾樣品獲取多次光譜的平均光譜建模，這在生產(chǎn)中是容易實(shí)現(xiàn)的。在P1、P2、P3位置各自透射光譜建立的PLS模型中，基于P2位置光譜建立的PLS模型最優(yōu)，校正相關(guān)系數(shù)為0.914，校正均方根誤差為0.380，預(yù)測(cè)相關(guān)系數(shù)為0.924，預(yù)測(cè)均方根誤差為0.404。若在線檢測(cè)時(shí)自定心裝置能將臍橙位置調(diào)整為果蒂水平，則檢測(cè)效果較好，但部分臍橙樣品的橫徑與縱徑相差不大（見表1），果形為類球形，目前生產(chǎn)應(yīng)用中的自定心裝置還無法自動(dòng)將臍橙果蒂全部調(diào)整為水平位置，因此無法達(dá)到預(yù)期效果，存在在線檢測(cè)部位與建模時(shí)采集光譜部位不一致的情況，影響檢測(cè)準(zhǔn)確性?；赑3位置光譜建立的PLS模型效果最差，校正相關(guān)系數(shù)為0.546，校正均方根誤差為0.794，預(yù)測(cè)相關(guān)系數(shù)為0.586，預(yù)測(cè)均方根誤差為0.799。該模型是通過果蒂向下位置透射光譜建立的，在本研究中采用的光源布置下，果蒂向下時(shí)果蒂與信號(hào)接收探頭正對(duì)，極大阻礙了光的穿透，因此該位置獲得的信號(hào)的信噪比較低，是影響臍橙SSC檢測(cè)準(zhǔn)確性的主要原因。

表4 基于CARS的PLS建模及預(yù)測(cè)結(jié)果

3 結(jié) 論

為了提高利用Vis-NIR透射光譜在線檢測(cè)臍橙內(nèi)部物質(zhì)含量的準(zhǔn)確性，該文利用特制的可見-近紅外透射光譜測(cè)量裝置采集了福本臍橙果蒂向上、水平、向下3個(gè)位置的透射光譜進(jìn)行試驗(yàn)研究。

1）在果蒂向上、水平、向下3個(gè)位置各自透射光譜建立的PLS模型中，基于果蒂水平放置時(shí)的透射光譜建立的PLS模型最優(yōu)，校正相關(guān)系數(shù)為0.914，校正均方根誤差為0.380，預(yù)測(cè)相關(guān)系數(shù)為0.924，預(yù)測(cè)均方根誤差為0.404。

2）基于果蒂向上、水平、向下3個(gè)位置平均透射光譜和加權(quán)透射光譜建立的PLS模型均取得了較好的預(yù)測(cè)結(jié)果，預(yù)測(cè)相關(guān)系數(shù)均大于0.91，預(yù)測(cè)均方根誤差均小于0.43。

今后將進(jìn)一步改進(jìn)和優(yōu)化試驗(yàn)裝置，提高臍橙SSC在線檢測(cè)的準(zhǔn)確性。

[1] Magwaza L S, Opara U L, Nieuwoudt H, et al. NIR spectroscopy applications for internal and external quality analysis of citrus fruit: A review[J]. Food and Bioprocess Technology, 2012, 5: 425－444.

[2] Xie L J, Wang A C, Xu H R, et al. Applications of near-infrared systems for quality evaluation of fruits: A-review[J]. Transactions of the ASABE, 2016, 59(2): 399－419.

[3] Nicolai B M, Defraeye T, Ketelaere B D, et al. Nondestructive measurement of fruit and vegetable quality[J]. Annual Review of Food Science and Technology, 2014, 5: 285－312.

[4] Magwaza L S, Opara U L, Nieuwoudt H, et al. NIR spectroscopy applications for internal and external quality analysis of citrus fruit-A review [J]. Food and Bioprocess Technology, 2012, 5: 425－444.

[5] Wang H L, Peng J Y, Xie C Q, et al. Fruit quality evaluation using spectroscopy technology: A review [J]. Sensors, 2015, 15: 11889－11927.

[6] Yuan L M, Cai J R, Sun L, et al. Nondestructive measurement of soluble solids content in apples by a portable fruit analyzer [J]. Food Analytical Methods, 2016, 9: 785－794.

[7] Fan S X, Guo Z M, Zhang B H, et al. Using Vis/NIR diffuse transmittance spectroscopy and multivariate analysis to predicate soluble solids content of apple [J]. Food Analytical Methods, 2016, 9: 1333－1343.

[8] Luo X, Ye Z Z, Xu H R, et al. Robustness improvement of NIR-based determination of soluble solids in apple fruit by local calibration [J]. Postharvest Biology and Technology, 2018, 139: 82－90.

[9] 李龍，彭彥昆，李永玉，等. 蘋果內(nèi)外品質(zhì)在線無損檢測(cè)分級(jí)系統(tǒng)設(shè)計(jì)與試驗(yàn)[J]. 農(nóng)業(yè)工程學(xué)報(bào), 2018, 34(9): 267－275. Li Long, Peng Yankun, Li Yongyu, et al. Design and experiment on grading system for online non-destructive detection of internal and external quality of apple[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(9): 267－275. (in Chinese with English abstract)

[10] 劉燕德，吳明明，孫旭東，等. 黃桃表面缺陷和可溶性固形物光譜同時(shí)在線檢測(cè)[J]. 農(nóng)業(yè)工程學(xué)報(bào), 2016, 32(6): 289－295. Liu Yande, Wu Mingming, Sun Xudong, et al. Simultaneous detection of surface deficiency and soluble solids content for amygdalus persica by online visible-near infrared transmittance spectroscopy[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2016, 32(6): 289－295. (in Chinese with English abstract)

[11] Nascimento P M, Carvalho L C, Junior L C, et al. Robust PLS models for soluble solids content and firmness determination in low chilling peach using near-infrared spectroscopy (NIR) [J]. Postharvest Biology and Technology, 2016, 111: 345－351.

[12] 介鄧飛，陳猛，謝麗娟，等. 適宜西瓜檢測(cè)部位提高近紅外光譜糖度預(yù)測(cè)模型精度[J]. 農(nóng)業(yè)工程學(xué)報(bào)，2014, 30(9): 229－234. Jie Dengfei, Chen Meng, Xie Lijuan, et al. Improving precision of soluble solid content predictive model by adopting suitable detective position of watermelon based on near infrared spectroscopy[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2014, 30(9): 229－234. (in Chinese with English abstract)

[13] 錢曼，黃文倩，王慶艷，等. 西瓜檢測(cè)部位差異對(duì)近紅外光譜可溶性固形物預(yù)測(cè)模型的影響[J]. 光譜學(xué)與光譜分析, 2016, 36(6): 1700－1705. Qian Man, Huang Wenqian, Wang Qingyan, et al. Assessment of influence detective position variability on precision of near infrared models for soluble solid content of watermelon [J]. Spectroscopy and Spectral Analysis, 2016, 36(6): 1700－1705. (in Chinese with English abstract)

[14] 樊書祥，黃文倩，李江波，等. LS-SVM 的梨可溶性固形物近紅外光譜檢測(cè)的特征波長篩選[J]. 光譜學(xué)與光譜分析，2014, 34(8): 2089－2093. Fan Shuxiang, Huang Wenqian, Li Jiangbo, et al. Characteristic wavelengths selection of soluble solids content of pear based on NIR spectral and LS-SVM[J]. Spectroscopy and Spectral Analysis, 2014, 34(8): 2089－2093. (in Chinese with English abstract)

[15] Sun X D, Liu Y D, Li Y F, et al. Simultaneous measurement of brown core and soluble solids content in pear by on-line visible and near infrared spectroscopy [J]. Postharvest Biology and Technology, 2016, 116: 80－87.

[16] Jamshidi B, Minaei S, Mohajerani E, et al. Reflectance Vis/NIR spectroscopy for nondestructive taste characterization of[J]. Computers and Electronics in Agriculture, 2012, 85: 64－69.

[17] Ncama K, Opara U L, Tesfay S Z, et al. Application of Vis/NIR spectroscopy for predicting sweetness and flavour parameters of() and() [J]. Journal of Food Engineering, 2017, 193: 86－94.

[18] Peiris K, Kays S. Spatial variability of soluble solids or dry-matter content within individual fruits, bulbs, or tubers implications for the development and use of NIR spectrometric techniques [J]. HortScience, 1999, 34(1): 114－117.

[19] 劉燕德，施宇，蔡麗君，等. 基于CARS算法的臍橙可溶性固形物近紅外在線檢測(cè)[J]. 農(nóng)業(yè)機(jī)械學(xué)報(bào), 2013, 44(9): 138－144. Liu Yande, Shi Yu, Cai Lijun, et al. On-line NIR detection model optimization of soluble solids content in navel orange based on CARS [J]. Transactions of Chinese Society for Agricultural Machinery, 2013, 44(9): 138－144. (in Chinese with English abstract)

[20] 劉燕德，胡軍，歐陽玉平，等. 贛南臍橙可溶性固形物近紅外光譜在線無損檢測(cè)[J]. 廣東農(nóng)業(yè)科學(xué), 2016, 43(9): 105－111. Liu Yande, Hu Jun, Ouyang Yuping, et al. Online detection of soluble solids content forby visible-near infrared diffuse transmission spectroscopy[J]. Guangdong Agricultural Sciences, 2016, 43(9): 105－111. (in Chinese with English abstract)

[21] Fraser D G, Jordan R B, Kunnemeyer R, et al. Light distribution inside mandarin fruit during internal quality assessment by NIR spectroscopy [J]. Postharvest Biology and Technology, 2003, 27: 185－196.

[22] 孫通，莫欣欣，劉木華. 果皮對(duì)臍橙可溶性固形物可見/近紅外檢測(cè)精度的影響[J]. 光譜學(xué)與光譜分析, 2018, 38(5): 1406－1411. Sun Tong, Mo Xinxin, Liu Muhua. Effect of pericarp on prediction accuracy of soluble solid content in navel oranges by visible/near infrared spectroscopy[J]. Spectroscopy and Spectral Analysis, 2018, 38(5): 1406－1411. (in Chinese with English abstract)

[23] 許文麗，孫通，吳文強(qiáng)，等. 臍橙放置位置對(duì)近紅外光譜檢測(cè)結(jié)果的影響[J]. 光譜學(xué)與光譜分析, 2012, 32(11): 3002－3005. Xu Wenli, Sun Tong, Wu Wenqiang, et al. Near-infrared spectrum detection result influenced by navel oranges placement position [J]. Spectroscopy and Spectral Analysis, 2012, 32(11): 3002－3005. (in Chinese with English abstract)

[24] Gomez A H, He Y, Pereira A G. Non-destructive measurement of acidity, soluble solids and firmness ofusing Vis/NIR spectroscopy techniques [J]. Journal of Food Engineering, 2006, 77: 313－319.

[25] Golic M, Walsh K, Lawson P. Short-wavelength near-infrared spectra of sucrose, glucose, and fructose with respect to sugar concentration and temperature [J]. Applied Spectroscopy, 2003, 57: 139－145.

[26] Magwaza L S, Opara U L, Terry L A, et al. Evaluation of fourier transform-NIR spectroscopy for integrated external and internal quality assessment of[J]. Journal of Food Composition and Analysis, 2013, 31: 144－154.

[27] Leardi R, Norgaard L. Sequential application of backward interval partial least squares and genetic of relevant spectral regions [J]. Journal of Chemometrics, 2004, 18: 486－497.

[28] Zhou X B, Zhao J W, Li Y X. Selection of the efficient wavelength regions in FT-NIR spectroscopy for determination of SSC ofbased on BiPLS and FiPLS models [J]. Vibrational Spectroscopy, 2007, 44: 220－227.

[29] Fan S X, Zhang B H, Li J B, et al. Effect of spectrum measurement position variation on the robustness of NIR spectroscopy models for soluble solids content of apple [J]. Biosystems Engineering, 2016, 143: 9－19.

[30] Li H D, Liang Y Z, Xu Q S, et al. Key wavelengths screening using competitive adaptive reweighted sampling method for multivariate calibration [J]. Analytica Chimica Acta, 2009, 648: 77－84.

[31] Dhanoa M S, Lister S J, Sanderson R, et al. The link between multiplicative scatter correction (MSC) and standard normal variate (SNV) transformations of NIR spectra [J]. Journal of Near Infrared Spectroscopy, 1994, 2: 43－47.

Detecting soluble solids content of navel orange based on transmission spectrum of tetragonal symmetric light source

Song Jie, Li Guanglin※, Yang Xiaodong, Zhang Xin, Liu Xuwen

(,,400715,)

Navel orange is a very popular fruit in China, which is mainly cultivated along the Yangtze River. Navel oranges are classified into different grades based on external quality and internal quality before they are sold. Soluble solids content is one of the main indices for evaluating the internal quality of navel orange. Therefore, it is very important to improve the detection accuracy of soluble solids content in production. So far, visible and near infrared spectroscopy (Vis-NIR) is one of the most widely used and effective techniques in internal quality assessment of fruits. In this study, 199 Fukumoto navel oranges were taken as experimental samples. The transmission spectra of navel oranges of three positions including pedicle upwards (P1), pedicle horizontal (P2) and pedicle downward (P3) were acquired by using a special visible and near infrared transmission spectrum measurement system designed by ourselves. The average spectra (P4) and weighted spectra (P5) of P1, P2 and P3 were calculated. The transmission spectra, including P1, P2, P3, P4 and P5 were preprocessed by multivariate scattering correction, standard normal variate transformation, first derivative and second derivative respectively. The best pretreatment results were obtained based on first derivative after comparative study. Then the spectra data preprocessed by first derivative were divided into 30 to 50 intervals with step length of 5, and backward interval partial least squares was used to select the optimal band combination. Good results observed when P1, P2, P3, P4 and P5 were divided into 35, 40, 30, 35 and 40 intervals, in which 161, 180, 114, 308 and 170 variables were retained. On this basis, competitive adaptive re-weighted sampling (CARS) was used to select feature variables. After running CARS for 20 times in each selection, 24, 23, 18, 39 and 22 variables were kept respectively. Finally, Five PLS models were established, including P1-PLS, P2-PLS, P3-PLS, P4-PLS and P5-PLS. Among the P1-PLS, P2-PLS and P3-PLS models, P2-PLS model was the best one, as the value of correlation coefficients of prediction was 0.924 and the value of root mean square error of predictionwas 0.404. This model can be realized by adjusting the navel oranges to pedicle horizontal in modeling. P4-PLS model and P5-PLS model had achieved good prediction results, as the value of correlation coefficients of prediction was higher than 0.91 and the value of root mean square error of prediction was lower than 0.43. P4-PLS model was based on the average spectra of P1, P2 and P3, and had potential to be realized by rolling the navel oranges in actual application. However, P5-PLS model was based on weighted spectra of P1, P2 and P3, which was difficult to realize in on-line detection. This study can provide a reference for the development of on-line detection equipment for the assessment of internal content of substances in navel orange.

fruit; spectrum analysis; models; navel orange; tetragonal symmetric light source; transmittance; soluble solids content

10.11975/j.issn.1002-6819.2019.10.034

S233.5; O657.33

A

1002-6819(2019)-10-0267-07

2018-12-27

2019-04-13

重慶市科委重點(diǎn)項(xiàng)目（cstc2018jszx-cyzdx0051）、中央高?；究蒲袠I(yè)務(wù)費(fèi)重點(diǎn)項(xiàng)目（XDJK2016B026）

宋 杰，講師，博士生，主要從事智能控制與檢測(cè)技術(shù)研究。Email：sj2008@swu.edu.cn

李光林，教授，博士生導(dǎo)師，主要從事傳感器與智能檢測(cè)技術(shù)研究。Email：liguanglin@swu.edu.cn

宋 杰，李光林，楊曉東，張 信，劉旭文. 基于四方對(duì)稱光源透射光譜的臍橙可溶性固形物檢測(cè)[J]. 農(nóng)業(yè)工程學(xué)報(bào)，2019，35(10)：267－273. doi：10.11975/j.issn.1002-6819.2019.10.034 http://www.tcsae.org

Song Jie, Li Guanglin, Yang Xiaodong, Zhang Xin, Liu Xuwen.Detecting soluble solids content of navel orange based on transmission spectrum of tetragonal symmetric light source[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2019, 35(10): 267－273. (in Chinese with English abstract) doi：10.11975/j.issn.1002-6819.2019.10.034 http://www.tcsae.org