徐 賽,陸華忠,王 旭,丘廣俊,王 陳,梁 鑫
基于可見/近紅外光譜的菠蘿水心病無損檢測
徐 賽1,陸華忠2※,王 旭1,丘廣俊1,王 陳1,梁 鑫1
(1. 廣東省農業科學院農業質量標準與監測技術研究所,廣州 510640; 2. 廣東省農業科學院,廣州 510640)
水心病近年嚴重危害菠蘿產業,探究一種菠蘿水心病的無損檢測方法對保證上市果品、指導采后處理、促進產業提升具有重要意義。該研究采用自行搭建的菠蘿可見/近紅外光譜無損智能檢測平臺,考慮實際應用成本與效果,搭載覆蓋不同波段(400~1 100、900~1 700和400~1 700 nm)的檢測器對菠蘿樣本進行采樣,隨后人工標定水心病發生程度。研究結果表明,3種不同光譜波段對菠蘿水心程度檢測的較優方法均為:采用全波段進行多項式平滑(Savitzky Golay,SG)處理,再進行標準正態變量校正(Standard Normal Variate,SNV),最后結合概率神經網絡(Probabilistic Neural Network,PNN)建模識別。其中,400~1 100 nm所建模型對菠蘿水心病訓練集的回判正確率為98.51%,對驗證集的檢測正確率為91.18%;900~1 700 nm所建模型對菠蘿水心病訓練集的回判正確率為100%,對驗證集的檢測正確率為62%;400~ 1 700 nm所建模型對菠蘿水心病訓練集的回判正確率為100%,對驗證集的檢測正確率為91.18%。主成分分析(Principal Component Analysis,PCA)和偏最小二乘回歸(Partial Least Squares Regression,PLSR)分析結果均顯示,采用400~ 1 700 nm能輕微提升400~1 100 nm的檢測效果。綜合考慮實際應用成本與效果,實際應用建議采用400~1 100 nm光譜結合SG + SNV + PNN對菠蘿水心病進行識別。研究結果證明可見/近紅外光譜技術可為菠蘿水心病無損、快速、智能檢測提供有效的解決方案,為相關領域提供參考。
無損檢測;模型;菠蘿;水心病;可見/近紅外光譜
水心病是菠蘿的生理性病害,過去時有發生,受關注較少,但近年中國菠蘿水心病發生逐年加重,成為產業的新問題[1]。發生水心病的菠蘿果肉呈腐爛、水浸狀,由于果肉細胞間隙充滿液體,這種果實不耐存放,并且會迅速散發出酒糟味和惡臭味,嚴重影響口感和風味,失去商品價值[2]。研究團隊2019-2021年對中國菠蘿主產區廣東徐聞菠蘿筆者調查的水心病發生率分別為15%、24%和44%,呈逐年遞增的趨勢,需引起相關領域重視。
田間水果品質的形成通常受陽光[3]、降雨[4]、氣溫[5]、營養[6]等諸多因素的影響,加上中國菠蘿以散戶種植為主,種植標準不統一,短期內想要根治水心病難度較大。因此,亟需一種無損、快速、有效的方法對水心菠蘿果實進行檢測與分級,指導采后處理、保障市場品質、保護品牌形象。據調研,目前產業普遍采用人工敲擊辯聲的方法識別,水心菠蘿果通常聲音較沉悶,但正確率只有約60%,且存在成本高、勞動強度大、檢測效率低。可見,開發一種無損、智能、快速菠蘿水心病檢測方法意義重大。
目前,可見/近紅外光譜[7]、電子鼻[8]和機器視覺[9]技術在農產品品質無損智能檢測中均發揮著重要作用。菠蘿水心病發生是從內部靠近果心的果眼位置開始,再逐漸向外蔓延。電子鼻和機器視覺技術在無損檢測過程中更側重于靠近農產品外表的特征,而可見/近紅外光可穿透農產品,獲取內部品質特征信息,更加適合于菠蘿水心病的無損智能檢測。前期研究表明,可見/近紅外光譜在小型薄皮水果的內部糖度[10-11]、酸度[12-13]、硬度[14-15]、病蟲害[16-19]等內部品質無損檢測上是可行的,但菠蘿屬于大型水果,且表面不光滑,容易引起散射噪聲,檢測難度相對較大[20]。采用可見/近紅外光譜技術能否有效無損檢測菠蘿水心病,尚未見有關報道。
為此,本研究基于可見/近紅外光譜技術自行搭建了一套菠蘿水心病無損智能檢測平臺,考慮實際應用成本與效果,基于平臺搭載覆蓋不同波段的檢測器對菠蘿樣本進行無損采樣,隨后切開檢測水心發生情況,建立菠蘿可見/近紅外光譜特征對水心病的無損檢測模型,為菠蘿產業開發水心病無損、快速、智能檢測方法提供科學參考。
自行搭建的菠蘿品質無損檢測實驗平臺如圖1所示。采樣時將菠蘿水平放置在載物臺的托盤上(托盤可固定菠蘿姿態,亦可使試驗結果更好地為流水線動態檢測提供參考)。為防止光線未經過菠蘿直接被光纖接收造成噪聲干擾,光源發射的光需經過進光孔,透射過樣本后,經過出光孔方可被接收。測試過程在暗箱內進行,箱體窗口用窗簾遮光。為尋找較優的菠蘿光譜采樣參數,平臺以下參數活動可調:光源0~900 W可調由9盞100 W的鹵素燈組成,LM-100型號,日本MORITEX公司,平均壽命為1 000 h),隔光板上進光孔與出光孔的大小經過多次更換、測試確定,光源、菠蘿樣本和接收光纖之間的距離可通過滑臺調節。
接收光纖另一端連接兩臺覆蓋不同波段的光譜儀,分別是QE pro和NIR QUESR(均為美國Ocean Optics公司生產),可覆蓋波段400~1 100和900~1 700 nm,若采用兩臺光譜儀聯用的方式可覆蓋400~1 700 nm的光譜信息。

1.光源 2.暗箱體 3.光源開關 4.隔光板 5.托盤 6.載物臺 7.遮光窗簾 8.光譜光纖 9.滑臺 10.出光孔 11.進光孔 12.菠蘿樣本
本試驗采用的菠蘿果實2021年4月采摘于廣東省湛江市徐聞縣某農場,品種為“巴厘”,共100個樣本,采果后立即在農場附件搭建的實驗房內進行采樣與測試。
經過調試,菠蘿可見/近紅外光譜的較優采集參數設置為:光譜儀QE pro與NIR QUEST的積分時間分別為 600 ms與2 000 ms;接收光纖距離菠蘿托盤距離30 mm;菠蘿托盤進光孔位置距離光源84 mm;光源為500 W;菠蘿托盤位于托盤的中心位置,光源、進光孔、菠蘿、出光孔、接收光纖處于同一水平。
采集菠蘿光譜信息后,立即進行水心病人工評判。目前尚未見菠蘿水心病評級方法,團隊前期研究提出[21]:將菠蘿縱切兩半,再平均切分成12小片平鋪在桌面上,較全面地觀察并記錄菠蘿水心病發生情況。無水心病表示無水心病發生,水心面積占總面積的0%;輕微水心病表示呈輕微水菠蘿跡象,仍可食用,具有一定商品價值,水心面積小于或等于總面積的10%;嚴重水心病表示果實水心病嚴重發生,無法食用,失去商品價值,水心面積大于總面積的10%。共采集到無水心病、輕微水心病、嚴重水心病樣本分別為56、21和23個。
采用主成分分析(Principal Component Analysis,PCA)[22]判別不同水心程度菠蘿的分類效果,由第一和第二主成分(The first and second principal component,PC1 and PC2)構成的樣本散點圖表示;采用多項式平滑(Savitzky Golay,SG)[23]濾波減少大型水果光譜采樣因光程較長、信噪比較低帶來的噪聲波動,濾波效果受多項式階次與平滑點數的影響;隨后采用標準正態變量校正(Standard Normal Variate,SNV)[24]降低菠蘿表皮極其粗糙等帶來的散射噪聲;SG + SNV預處理后,采用連續投影算法(Successive Projections Algorithm, SPA)[25]+ PCA + 歐氏距離(Euclidean Distance,ED)[26]進行光譜特征提取,其中SPA根據差異大小進行光譜特征的排序,特征數量從2到最大逐漸增加,分別進行PCA處理,采用ED計算不同類別中心點之間的距離,以距離的大小判斷增加特征的必要性;最后,對預處理與特征提取后的光譜數據,采用偏最小二乘回歸(Partial Least Squares Regression,PLSR)[27]與概率神經網絡(Probabilistic Neural Network,PNN)[28]分訓練集與校正集進行進一步建模判別,無、輕度和重度水心病分別隨機選擇38、14和15個樣本作為訓練集,其余19、7和8個樣本作為驗證集,不同水心程度由小到大期望輸出均分別設定為1、2和3,其中PLSR的檢測效果受降維后特征個數的選取影響較大,結果輸出為小數,通常用預測值與實際值之間的決定系數2,以及均方根誤差(Root Mean Square Error, RMSE)表示,PNN的檢測效果受擴散速度Spread值影響較大,其結果輸出為整數,可直接用正確率表達。為進一步統計PLSR的識別正確率,將PLSR結果輸出進行四舍五入取整,小于等于1的結果輸出為無水心,等于2為輕微水心,大于等于3為重度水心。
2.1.1 原始數據+PCA判別
菠蘿樣本在400~1 100 nm的原始光譜如圖2a所示,數據在1 000 nm以后出現輕微的噪聲波動。400~1 100 nm原始數據對菠蘿水心程度的PCA判別結果如圖2b所示。不同水心程度菠蘿樣本可以被區分開來,但距離較近,且離散程度較高,聚類性較差。
菠蘿樣本在900~1 700 nm的原始光譜如圖3a所示,數據均存在明顯的噪聲波動,且隨波長增加而增大。900~1 700 nm原始數據對菠蘿水心程度的PCA判別結果如圖3b所示。不同水心程度菠蘿樣本無法被區分開來。
菠蘿樣本在400~1 700 nm的原始光譜如圖4a所示,數據在1 000 nm以后噪聲波動逐漸增強。400~1 700 nm原始數據對菠蘿水心程度的PCA判別結果如圖4b所示。第一主成分(PC1)與第二主成分(PC2)的貢獻率分別為60.77和32.59%,總貢獻率為93.36%。與400~1 100 nm光譜分類結果圖相似(圖2b),不同水心程度菠蘿樣本可以被區分開來,但距離較近,離散程度較高,聚類性較差。
2.1.2 SG濾波+SNV校正+PCA判別
為提高光譜數據質量,經試驗,采用3階23點SG處理可較好地濾除400~1 100 nm光譜數據中存在的噪聲波動,隨后采用SNV對光譜信號中的散射噪聲進行校正,得到處理后的菠蘿光譜信號如圖5a所示。基于處理后的光譜信號對菠蘿水心程度進行PCA判別的結果如圖5b所示。對比圖2b,PCA同樣可以有效區分不同水心程度,且同類樣本數據點的聚類性明顯增強,但不同樣本之間存在少量數據點重疊,實際分類中有誤判的風險。
為提高光譜數據質量從而提升檢測效果,經反復試驗,采用3階41點SG處理可較好地濾除900~1 700 nm光譜數據中存在的噪聲波動,隨后采用SNV對光譜信號中的散射噪聲進行校正,得到處理后的菠蘿光譜信號如圖6a所示。基于處理后的光譜信號對菠蘿水心程度進行PCA判別的結果如圖6b所示。PCA無法有效區分不同水心程度,但對比圖3b,樣本數據點的聚類性明顯增強。
為保障整體光譜曲線的銜接性與降噪效果,采用3階41點SG處理并濾除400~1 700 nm光譜數據中存在的噪聲波動,隨后采用SNV對光譜信號中的散射噪聲進行校正,得到處理后的菠蘿光譜信號如圖7a所示。處理后的光譜信號對菠蘿水心程度進行PCA判別的結果如圖7b所示。PCA同樣可以有效區分不同水心程度,對比圖4b,重疊的數據點個數略有減少,但聚類性略有降低,部分樣本實際分類中仍有誤判的風險。
2.1.3 SPA+PCA+ED特征提取
為明確是否每一個特征對分類識別均有積極作用,采用SPA + PCA + ED對400~1 100 nm光譜特征作用的分析結果如圖8a所示。采用SPA將特征作用從大到小進行排序后,按順序逐漸增加特征數量并進行PCA分析,不同水心程度數據點之間的ED逐漸增加。可見,400~ 1 100 nm所有的特征在分類識別過程中均是有益的。
采用SPA + PCA + ED對900~1 700 nm光譜特征作用的分析結果如圖8b所示。采用SPA將特征作用從大 到小進行排序后,按順序逐漸增加特征數量并進行PCA分析,不同水心程度數據點之間的歐式距離ED逐漸增加。可見,900~1700 nm所有的特征在分類識別過程中均是有益的。
采用SPA + PCA + ED對400~1700 nm光譜特征作用的分析結果如圖8c所示。采用SPA將特征作用從大到小進行排序后,按順序逐漸增加特征數量并進行PCA分析,不同水心程度數據點之間的ED逐漸增加。該結果進一步證明,400~1 700 nm所有的特征在分類識別過程中均是有益的。
2.1.4 PLSR、PNN檢測建模
為進一步探究可見/近紅外光譜對水心病無損檢測的應用效果,分別采用PLSR和PNN結合預處理與特征提取后的不同波段光譜進行檢測,結果如表1所示。
采用PLSR結合預處理與特征提取后的400~1 100 nm光譜數據分訓練集與驗證集對菠蘿水心病進行檢測,經反復訓練,PLSR的建模參數FN設定為11,模型對訓練集的PLSR回判R2和RMSEC分別為0.95與0.18,對于驗證集的檢測2和RMSEV分別為0.81和0.37,400~1 100 nm光譜對訓練集的回判正確率為98.51%(1個重度水心誤判為輕度水心),對測試集的檢測正確率為88.24%(1個輕度水心誤判為無水心;3個重度水心誤判為輕度水心)。采用PLSR結合預處理與特征提取后的900~1 700 nm光譜數據分訓練集與驗證集對菠蘿水心病進行檢測,經反復訓練,PLSR的建模參數FN設定為11,模型對訓練集的PLSR回判R2和RMSEC分別為0.76與0.40,對于驗證集的檢測2和RMSEV分別為0.45和0.62,對訓練集的回判正確率為80.60%(無水心中4個誤判為輕度水心;輕度水心中3個誤判為無水心,1個誤判為重度水心;重度水心中5個誤判為輕度水心),對測試集的檢測正確率為58.82%(無水心中5個誤判為輕度水心;輕度水心中3個誤判為無水心;重度水心中6個誤判為輕度水心),效果不佳。采用PLSR結合預處理與特征提取后的400~1700 nm光譜數據分訓練集與驗證集對菠蘿水心病進行檢測,經反復訓練,PLSR的建模參數FN設定為14,模型對訓練集的PLSR回判R2和RMSEC分別為0.96與0.17,對于驗證集的檢測2和RMSEV分別為0.83和0.35,對訓練集的回判正確率為100%,對測試集的檢測正確率為88.24%(3個無水心誤判為輕度水心;1重度水心誤判為輕度水心)。采用PNN結合預處理與特征提取后的400~1 100 nm光譜數據分訓練集與驗證集對菠蘿水心病進行建模檢測,經反復訓練,PNN模型參數Spread設定為1.2,所建模型對訓練集的回判正確率為98.51%(1個重度水心誤判為輕度水心),對驗證集的檢測正確率為91.18%(1個輕度水心誤判為無水心;2個重度水心誤判為輕度水心),具有較好地檢測效果。采用PNN結合預處理與特征提取后的900~1700 nm光譜數據分訓練集與驗證集對菠蘿水心病進行建模檢測,經反復訓練,PNN模型參數Spread設定為0.1,所建模型對訓練集的回判正確率為100%,對驗證集的檢測正確率為62%(無水心中1個誤判為輕度水心,4個誤判為重度水心;輕度水心中4個誤判為無水心,1和誤判為無水心;重度水心中1個誤判為輕度水心,2個誤判為無水心),檢測效果不佳。采用PNN結合預處理與特征提取后的400~1 700 nm光譜數據分訓練集與驗證集對菠蘿水心病進行建模檢測,經反復訓練,PNN模型參數Spread設定為0.2,所建模型對訓練集的回判正確率為100%,對驗證集的檢測正確率為91.18%(1個輕度水心誤判為無水心;2個重度水心誤判為輕度水心),具有較好地檢測效果。

表1 不同波段對菠蘿水心病的檢測精度與成本
注:FN為偏最小二乘模型的特征因子數,Spread為概率神經網絡模型的散布常數。
Note: FN is the feature factor number of PLSR model, Spread is the spread constant of PNN model.
菠蘿水心病的發生伴隨著果肉質地、顏色以及成分等變化,對其他小型薄皮水果前期研究表明[29-30],這些特征均可被可見/近紅外光譜捕獲,因此,本文采用可見/近紅外光譜檢測菠蘿水心病發生程度是有依據支撐的。本文進一步驗證了可見/近紅外光譜結合信號預處理以及模式識別,無損檢測菠蘿內部水心病發生程度是可行的。
菠蘿屬于大型水果,檢測時光的譜透過性較差,造成信號波動,且表面極為粗糙,易形成散射噪聲。因此,本文采用SG與SNV處理可有效降低信號波動以及散射噪聲來帶的干擾,提升識別效果。特征提取主要在于剔除會降低識別精度的噪聲,最大化地保留有益信息形成信息融合,本文提出采用SPA + PCA + ED分析結果表明,所有特征均包含分類識別的有益信息,均應保留。
QE pro(400~1 100 nm)比NIR QUEST(900~1 700 nm)具有更好的檢測效果,是因為400~1 100 nm同時對質地、顏色以及成分變化敏感,而900~1 700 nm僅對質地和成分變化敏感[31]。此外,波長越長,光能越低,加上近紅外波段的光易被水果中的水分吸收,使得通過樣本后衰減較大,信噪比較低[32]。PLSR結果表明,采用QE pro與NIR QUEST聯用(400~1 700 nm)可略微提升QE pro的檢測效果,是因為1100~1 700 nm包含菠蘿水心病識別的有益信息,可對400~1 700 nm形成信息補充與融合[33],但該方式增加檢測成本較大,性價比較低。實際應用建議單獨采用400~1 700 nm進行菠蘿水心病檢測。
PCA對菠蘿水心病程度的分類結果可以看出,不同類別樣本數據點之間不能用一條直線完全劃分開來,存在一定非線性特性。而PNN和PLSR的映射方式分別是神經網絡和線性回歸,即PNN比PLSR的識別運算函數具有更強的非線性分類識別能力。因此,PNN在解決菠蘿水心病發生程度的檢測上具有更好的檢測效果。
1)采用400~1 100 nm光譜原數據結合主成分分析(Principal Component Analysis,PCA)分析可將不同水心程度菠蘿樣本區分開來,但距離較近,且離散程度較高,聚類性較差。采用900~1 700 nm光譜原數據結合PCA分析無法將不同水心程度菠蘿樣本區分開來。采用400~1 700 nm光譜原數據結合PCA分析可將不同水心程度菠蘿樣本區分開來,相對400~1 100 nm的檢測效果略有提高。
2)經多項式平滑(Savitzky Golay,SG) + 標準正態變量校正(Standard Normal Variate,SNV)處理400~1 100 nm光譜后,PCA同樣可以有效區分不同水心程度,且同類樣本數據點的聚類性明顯增強,但不同樣本之間存在少量數據點重疊,存在誤判的風險。經SG + SNV處理900~1700 nm光譜后,PCA分析對樣本數據點的聚類性明顯增強,但分類效果仍不佳。經SG + SNV處理400~1 700 nm光譜后,可增強同類樣本數據點的聚類性。連續投影算法(Successive Projections Algorithm, SPA)+ (Principal Component Analysis,PCA)+歐氏距離(Euclidean Distance,ED)分析結果顯示,400~1 100 nm、900~1 700 nm、400~1 700 nm 3種波段選擇包含的特征在分類識別過程中均是有益的,均應被保留。
4)偏最小二乘回歸(Partial Least Squares Regression,PLSR)結合400~1 100 nm光譜數據所建模型對菠蘿水心病訓練集的回判正確率為98.51%,對測試集的檢測正確率為88.24%。PLSR結合900~1 700 nm光譜數據所建模型對菠蘿水心病訓練集的回判正確率為80.60%,對測試集的檢測正確率為58.82%。PLSR結合400~1 700 nm光譜數據所建模型對菠蘿水心病訓練集的回判正確率為100%,對測試集的檢測正確率為88.24%。概率神經網絡(Probabilistic Neural Network,PNN)結合400~1 100 nm光譜數據所建模型對菠蘿水心病訓練集的回判正確率為98.51%,對驗證集的檢測正確率為91.18%。PNN結合900~1 700 nm光譜數據所建模型對菠蘿水心病訓練集的回判正確率為100%,對驗證集的檢測正確率為62%。PNN結合400~1700 nm光譜數據所建模型對菠蘿水心病訓練集的回判正確率為100%,對驗證集的檢測正確率為91.18%。
5)綜合考慮成本與效果,實際應用建議采用400~ 1 100 nm光譜結合多項式平滑(Savitzky Golay,SG) +標準正態變量校正(Standard Normal Variate,SNV) +概率神經網絡(Probabilistic Neural Network,PNN)對菠蘿水心病進行識別。下一步研究一方面可進一步提出信號處理新方法,減少建模特征數量,簡化模型,另一方面可運用模型對大批量菠蘿進行試驗驗證,不斷修正模型參數以提高模型適應性,更好地服務產業。
[1] 賀軍虎,欒愛萍. 菠蘿水心病和裂果的發生與防治措施[J]. 草坪常見病蟲害,2019,7:41.
[2] Shu H Y, Sun W, Li K M, et al. The cause for water-heart fruit of pineapple and protective measurements[J]. American Journal of Plant Sciences, 2019, 10(6): 885-892.
[3] Kishore K, Rupa T R, Samant D. Influence of shade intensity on growth, biomass allocation, yield and quality of pineapple in mango-based intercropping system[J]. Scientia Horticulturae, 2021, 278: 109868.
[4] Williams P A, Crespo O, Atkinson C J, et al. Impact of climate variability on pineapple production in Ghana[J]. Agriculture & Food Security, 2017, 6(1): 1-14.
[5] Friend D J C. Effect of night temperature on flowering and fruit size in pineapple ([L.] Merrill)[J]. Botanical Gazette, 1981, 142(2): 188-190.
[6] Liu C H, Liu Y, Fan C, et al. The effects of composted pineapple residue return on soil properties and the growth and yield of pineapple[J]. Journal of Soil Science and Plant Nutrition, 2013, 13(2): 433-444.
[7] Pahlawan M F R, Wati R K, Masithoh R E. Development of a low-cost modular VIS/NIR spectroscopy for predicting soluble solid content of banana[C]// IOP Conference Series: Earth and Environmental Science, Banda Aceh: IOP Publishing, 2021, 644: 012047.
[8] Shi H, Zhang M, Adhikari B. Advances of electronic nose and its application in fresh foods: A review[J]. Critical Reviews in Food Science and Nutrition, 2018, 58(16): 2700-2710.
[9] 張思雨,張秋菊,李可. 采用機器視覺與自適應卷積神經網絡檢測花生仁品質[J]. 農業工程學報,2020,36(4)::269-277.
Zhang Siyu, Zhang Qiuju, Li Ke. Detection of peanut kernel quality based on machine vision and adaptive convolution neural network[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2020, 36(4): 269-277. (in Chinese with English abstract)
[10] 邵園園,王永賢,玄冠濤,等. 高光譜成像快速檢測殼聚糖涂膜草莓可溶性固形物[J]. 農業工程學報,2019,35(18):245-254.
Shao Yuanyuan, Wang Yongxian, Xuan Guantao, et al. Rapid detection of soluble solids content in strawberry coated with chitosan based on hyperspectral imaging[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2019, 35(18): 245-254. (in Chinese with English abstract)
[11] Yang B, Guo W C, Li W Q, et al. Portable, visual, and nondestructive detector integrating Vis/NIR spectrometer for sugar content of kiwifruits[J]. Journal of Food Process Engineering, 2019, 42(2): e12982.
[12] Ncama K, Opara U L, Tesfay S Z, et al. Application of Vis/NIR spectroscopy for predicting sweetness and flavour parameters of ‘Valencia’ orange () and ‘Star Ruby’ grapefruit (Citrus × paradisi Macfad)[J]. Journal of Food Engineering, 2017, 193: 86-94.
[13] Jamshidi B, Minaei S, Mohajerani E, et al. Reflectance Vis/NIR spectroscopy for nondestructive taste characterization of Valencia oranges[J]. Computers and Electronics in Agriculture, 2012, 85: 64-69.
[14] Uwadaira Y, Sekiyama Y, Ikehata A. An examination of the principle of non-destructive flesh firmness measurement of peach fruit by using VIS-NIR spectroscopy[J]. Heliyon, 2018, 4(2): e531.
[15] Yu X J, Lu H D, Wu D. Development of deep learning method for predicting firmness and soluble solid content of postharvest Korla fragrant pear using Vis/NIR hyperspectral reflectance imaging[J]. Postharvest Biology and Technology, 2018, 141: 39-49.
[16] Sun X D, Zhu K, Jiang X G, et al. Non-destructive detection of blackheart and soluble solids content of intact pear by online NIR spectroscopy[J]. The Journal of Supercomputing, 2020, 76(5): 3173-3187.
[17] Mogollon M R, Jara A F, Contreras C, et al. Quantitative and qualitative VIS-NIR models for early determination of internal browning in ‘Cripps Pink’apples during cold storage[J]. Postharvest Biology and Technology, 2020, 161: 111060.
[18] Jamshidi B. Ability of near-infrared spectroscopy for non-destructive detection of internal insect infestation in fruits: Meta-analysis of spectral ranges and optical measurement modes[J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2020, 225: 117479.
[19] Botu, M M, Bolvansky D, Cecchini B E, et al. NIR spectroscopy is suitable to detect insect infested chestnuts[J]. Acta Horticulturae, 2014, 1043: 153-159.
[20] 徐賽,陸華忠,丘廣俊,等. 水果品質無損檢測研究進展及應用現狀[J]. 廣東農業科學,2020,47(12):229-236.
Xu Sai, Lu Huazhong, Qiu Guangjun, et al. Research progress and application status of fruit quality nondestructive detection technology[J]. Guangdong Agricultural Sciences, 2020, 47(12): 229-236. (in Chinese with English abstract)
[21] 徐賽,梁鑫,陸華忠,等. 一種菠蘿水心病評級裝備: 213181273U [P]. 2020-12-18.
[22] Po?ízka P, Klus J, Képe? E, et al. On the utilization of principal component analysis in laser-induced breakdown spectroscopy data analysis, a review[J]. Spectrochimica Acta Part B: Atomic Spectroscopy, 2018, 148: 65-82.
[23] Jahani S, Setarehdan S K, Boas D A, et al. Motion artifact detection and correction in functional near-infrared spectroscopy: A new hybrid method based on spline interpolation method and Savitzky–Golay filtering[J]. Neurophotonics, 2018, 5(1): 15003.
[24] Bi Y M, Yuan K L, Xiao W Q, et al. A local pre-processing method for near-infrared spectra, combined with spectral segmentation and standard normal variate transformation[J]. Analytica chimica acta, 2016, 909: 30-40.
[25] Krepper G, Romeo F, Fernandes D D, et al. Determination of fat content in chicken hamburgers using NIR spectroscopy and the Successive Projections Algorithm for interval selection in PLS regression (iSPA-PLS)[J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2018, 189: 300-306.
[26] He W, Zhou J, Cheng H, et al. Validation of origins of tea samples using partial least squares analysis and Euclidean distance method with near-infrared spectroscopy data[J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2012, 86: 399-404.
[27] Zhu Z H, Li J M, Guo Y M, et al. Accuracy improvement of boron by molecular emission with a genetic algorithm and partial least squares regression model in laser-induced breakdown spectroscopy[J]. Journal of Analytical Atomic Spectrometry, 2018, 33(2): 205-209.
[28] Huang X D, Wang C Y, Fan X M, et al. Oil source recognition technology using concentration-synchronous- matrix-fluorescence spectroscopy combined with 2D wavelet packet and probabilistic neural network[J]. Science of the Total Environment, 2018, 616: 632-638.
[29] Alhamdan A M, Fickak A, Atia A R. Evaluation of sensory and texture profile analysis properties of stored Khalal Barhi dates nondestructively using Vis/NIR spectroscopy[J]. Journal of Food Process Engineering, 2019, 42(6): e13215.
[30] Arendse E, Fawole O A, Magwaza L S, et al. Fourier transform near infrared diffuse reflectance spectroscopy and two spectral acquisition modes for evaluation of external and internal quality of intact pomegranate fruit[J]. Postharvest Biology and Technology, 2018, 138: 91-98.
[31] 梁亮,劉志霄,楊敏華,等. 基于可見/近紅外反射光譜的稻米品種與真偽鑒別[J]. 紅外與毫米波學報,2009,28(5):353-356.
Liang Liang, Liu Zhixiao, Yang Minhua, et al. Discrimination of variety and authenticity for rice based on visual/near infrared reflection spectra[J].Journal of Infrared and Millimeter Waves, 2009, 28(5): 353-356. (in Chinese with English abstract)
[32] 劉臻臻,陳少藩,余彤泳,等. 自然光照對眼部的影響[J]. 國際眼科雜志,2020,20(3):191-196.
Liu Zhenzhen, Chen Shaofan, Yu Tongyong, et al. Effects of sunlight on eye[J].International Eye Science. 2020, 20(3): 191-196. (in Chinese with English abstract)
[33] Li H, Huang H H, Li Y F, et al. Physics of failure-based reliability prediction of turbine blades using multi-source information fusion[J]. Applied Soft Computing, 2018, 72: 624-635.
Nondestructive detection method for pineapple water core based on visible/near infrared spectroscopy
Xu Sai1, Lu Huazhong2※, Wang Xu1, Qiu Guangjun1, Wang Chen1, Liang Xin1
(1.,510640,; 2.,510640,)
Water core is a serious physiological disorder of pineapple in recent years. Effective detection of internal water core is highly urgent for the market quality of pineapple after post-harvest treatments. In this study, A nondestructive detection platform was lab-developed for the water core of pineapple usingVisible/Near-infrared (VIS/NIR) spectroscopy. The optimal parameters of the platform were set, where the integral time of 400-1 100 nm and 900-1 700 nm spectrometer were 600 and 2 000 ms, respectively, the intensity of light source was 500 W, the distance between the optical fiber and tray was 30 mm, the distance between the tray and input optical hole was 84 mm, while, all the light, input optical hole, pineapple sample, output optical hole, and optical fiber were in the same horizontal line. Three settings of spectrum wavelength (400-1 100 nm VIS/NIR spectrum, 900-1 700 nm NIR spectrum, and 400-1 700 nm VIS/NIR spectrum) were applied for the pineapple sampling. After that, the pineapple was cut open to artificially and immediately record the water core. The Savitzky Golay (SG) and Standard Normal Variate (SNV) were also applied for the subsequent data processing. Furthermore, the extraction of the feature was conducted using the Successive Projections Algorithm (SPA), Principal Component Analysis (PCA), and Euclidean Distance (ED). Some models were finally established using the Partial Least Squares Regression (PLSR) and Probabilistic Neural Network (PNN). The results showed that an optimal procedure of detection was achieved for the water core using three settings of spectrum wavelength: to take the full wavelength data for SG and SNV processing, and then build a detection model by PNN. Using 400-1 100 nm spectrum and the optimal detection, the accuracy of the model for the calibration set of the water core was 98.51%, while the accuracy of the model for the validation set was 91.18%. Using 900-1 700 nm spectrum and the optimal detection, the accuracy of the model for the calibration set of the water core was 100%, while, the accuracy of the model for the validation set was 62%. Using 400-1 700 nm spectrum and the optimal detection, the accuracy of the model for the calibration set of water core was 100%, while the accuracy of the model for the validation set was 91.18%. Besides, both PCA and PLSR showed that there was a relatively less significant improvement, even though the detection of water core was slightly improved by 400-1 700 nm spectrum, compare with only by 400-1 100 nm. Thus, a practical detection of water core was suggested to use the 400-1 100 nm spectrum that combined with SG + SNV + PNN modeling in industrial production. Specifically, the marking price of 400-1 100 nm spectrometer like QE pro was about 130 000 Yuan, and the marking price of 900-1 700 nm spectrometer like NIR QUEST was about 150 000 Yuan, while, the marking price of 400-1 700 nm spectrometer like a combination of QE pro and NIR QUEST was about 280 000 Yuan. Consequently, the VIS/NIR spectroscopy can be widely expected to nondestructively and rapidly identify the internal water core of pineapple in modern agriculture.
nondestructive detection; models; pineapple; water core; visible/near infrared spectroscopy
10.11975/j.issn.1002-6819.2021.21.033
TP29
A
1002-6819(2021)-21-0287-08
徐賽,陸華忠,王旭,等. 基于可見/近紅外光譜的菠蘿水心病無損檢測[J]. 農業工程學報,2021,37(21):287-294.doi:10.11975/j.issn.1002-6819.2021.21.033 http://www.tcsae.org
Xu Sai, Lu Huazhong, Wang Xu, et al. Nondestructive detection method for pineapple water core based on visible/near infrared spectroscopy[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2021, 37(21): 287-294. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2021.21.033 http://www.tcsae.org
2021-06-22
2021-08-10
廣東省鄉村振興戰略專項(403-2018-XMZC-0002-90);廣東省自然科學基金項目(2021A1515010834);國家自然科學基金項目(31901404);廣東省農業科學院十四五新興學科團隊建設項目(202134T);廣東省農業科學院金穎之星人才培養項目(R2020PY-JX020);廣東省農業科學院創新基金項目(202034)
徐賽,博士,副研究員,研究方向為農產品品質無損檢測技術與裝備。Email:xusai@gdaas.cn
陸華忠,博士,教授,博士生導師,研究方向為農產品物流保鮮與智能檢測技術。Email:huazlu@scau.edu.cn