李鑫星,朱晨光,周 婧,孫龍清,曹霞敏,張小栓
?
光譜技術(shù)在水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測中的應(yīng)用進展及趨勢
李鑫星1,2,朱晨光1,周 婧1,孫龍清1,曹霞敏3,張小栓2,4※
(1. 中國農(nóng)業(yè)大學(xué)信息與電氣工程學(xué)院,北京 100083;2. 食品質(zhì)量與安全北京實驗室,北京 100083; 3. 蘇州大學(xué)基礎(chǔ)醫(yī)學(xué)與生物科學(xué)學(xué)院,蘇州 215200;4. 中國農(nóng)業(yè)大學(xué)工學(xué)院,北京 100083)
水產(chǎn)養(yǎng)殖的水質(zhì)是關(guān)乎水產(chǎn)養(yǎng)殖經(jīng)濟效益和水產(chǎn)品品質(zhì)的關(guān)鍵因素,與傳統(tǒng)的水質(zhì)檢測方法相比,光譜技術(shù)具有無創(chuàng)性、快速性、可重復(fù)性、準(zhǔn)確性等優(yōu)點,已成為水質(zhì)監(jiān)測的重要發(fā)展方向。該文總結(jié)和整理現(xiàn)有國內(nèi)外研究文獻,對基于光譜技術(shù)的水質(zhì)重要參數(shù)監(jiān)測、數(shù)據(jù)預(yù)處理方法、特征波段提取、預(yù)測模型算法進行了系統(tǒng)的分析與討論。綜述結(jié)果表明,實時在線的水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測將成為重點研究方向;多源光譜融合、多參數(shù)的水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測將會成為新的發(fā)展方向;對于光譜數(shù)據(jù)的處理,將多種數(shù)據(jù)處理算法相結(jié)合,仍將占據(jù)主導(dǎo);而非線性建模將成為水產(chǎn)養(yǎng)殖水質(zhì)數(shù)據(jù)分析的主流方法非線性數(shù)據(jù)建模,將成為光譜技術(shù)應(yīng)用于水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測的主流建模發(fā)方法。
光譜技術(shù);水產(chǎn)養(yǎng)殖;水質(zhì);監(jiān)測模型
水產(chǎn)養(yǎng)殖已經(jīng)成為中國發(fā)展最快的食品生產(chǎn)行業(yè)之一,為保障食物供給、促進經(jīng)濟增長做出了巨大貢獻。水產(chǎn)養(yǎng)殖與其水質(zhì)密切相關(guān)[1],近年來,隨著經(jīng)濟的發(fā)展,工業(yè)廢水、生活污水的排放量大增,造成環(huán)境污染,養(yǎng)殖池塘水質(zhì)遭到污染的情況時有發(fā)生。作為智能農(nóng)業(yè)和農(nóng)業(yè)物聯(lián)網(wǎng)的重要研究內(nèi)容,水產(chǎn)養(yǎng)殖水質(zhì)信息的快速、準(zhǔn)確獲取,以求在環(huán)保、節(jié)能的同時達到高產(chǎn)、安全養(yǎng)殖的目的,成為學(xué)者們關(guān)心的問題?;诠庾V分析的水質(zhì)監(jiān)測技術(shù)是水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測的一個重要發(fā)展方向,與傳統(tǒng)的化學(xué)分析、電化學(xué)分析和色譜分析等方法相比,光譜分析技術(shù)更具有操作簡便、消耗試劑量小、重復(fù)性好、測量精度高和檢測快速的優(yōu)點,非常適合對水質(zhì)的快速在線監(jiān)測。本文綜述國內(nèi)外光譜技術(shù)在水產(chǎn)養(yǎng)殖水質(zhì)指標(biāo)快速監(jiān)測方面的應(yīng)用,總結(jié)并展望其未來發(fā)展。
水產(chǎn)養(yǎng)殖水質(zhì)參數(shù)變化將直接影響水產(chǎn)品的生長,對于水產(chǎn)養(yǎng)殖業(yè)來說,水體溶解氧pH值、水溫對水中生物的生存有著至關(guān)重要的影響。不同的養(yǎng)殖環(huán)境和養(yǎng)殖對象,對水質(zhì)參數(shù)的要求不同。針對魚類,水質(zhì)指標(biāo)控制范圍如下,pH值:淡水6.5~8.5,海水7.0~8.5。溶解氧連續(xù)24 h中,16 h以上必須大于5 mg/L,其余任何時候不得低于3 mg/L;氮元素,氨氮含量要低于0.2 mg/L,凱氏氮不高于0.05 mg/L,亞硝酸鹽低于0.1 mg/L,非離子氨不高于0.02 mg/L;對于磷元素,黃磷不高于0.001 mg/L;重金屬,汞不高于0.000 5 mg/L,鉻不高于0.005 mg/L,鉛不高于0.05 mg/L,銅不高于0.1 mg/L。
化學(xué)需氧量,簡稱COD,是指在一定條件下,水體中還原性物質(zhì)被強氧化劑氧化時,所消耗的氧化劑的量,是表征水中還原性物質(zhì)的綜合性指標(biāo)。COD是評價水質(zhì)極為重要的指標(biāo),它被用來衡量水體受還原性物質(zhì)污染的程度,是水質(zhì)檢測時必須要檢測的參數(shù)[2]。很多相關(guān)研究表明當(dāng)水體中COD的濃度超過一定的限值時,會對水產(chǎn)品的生長造成影響,而且會增加水產(chǎn)養(yǎng)殖廢水的處理成本。檢測COD的常規(guī)方法主要是高錳酸鹽指數(shù)法(CODMn)和重鉻酸鉀回流法(CODCr)。兩者的適用范圍不同,重鉻酸鉀法適用于生活廢水和工業(yè)廢水的測定。而高錳酸鉀法更適用于清潔的水質(zhì),這些方法存在操作復(fù)雜、耗時長、消解時易造成附加污染等問題[3]。
氮是水體中的主要營養(yǎng)物質(zhì)之一,水環(huán)境中氮的形態(tài)有氨氮、硝酸鹽氮、亞硝酸鹽氮、有機氮和總氮,前四者通過生物化學(xué)作用可以相互轉(zhuǎn)化??偟獮榍八恼咧停呛饬克w受污染程度的重要指標(biāo)[4]??偟獫舛葯z測方法主要有離子色譜法、化學(xué)滴定法、流動注射法、離子選擇電極法以及光譜分析法等,其中,化學(xué)滴定法的分析精度最高,但此類方法過程繁復(fù),耗時長,不適宜大范圍使用[5]。
總磷是衡量水質(zhì)的重要指標(biāo),也是評定水質(zhì)富營養(yǎng)化的重要指標(biāo)。在水體中磷類物質(zhì)含量過大會造成藻類過度繁殖,使水透明度降低,水質(zhì)變差,從而影響水產(chǎn)養(yǎng)殖產(chǎn)品的品質(zhì)。目前,中國總磷檢測一般按照原國家環(huán)保部發(fā)布的鉬酸銨分光光度法進行,國內(nèi)外用堿性過硫酸鉀消解—離子色譜法、過硫酸鉀消解法、硝酸—硫酸消解法、硝酸—高氯酸消解法測量水質(zhì)中的總磷也有報道[6-7]。
重金屬是水環(huán)境中較為危險的污染物,不僅不可降解,而且會在生物體內(nèi)長期積累,引起多種疾病。常用的檢測方法包括:原子吸收光譜法、電感耦合等離子體原子發(fā)射光譜法、電化學(xué)方法、紫外-可見分光光度法、液相色譜法、熒光分析法、流動注射分析、生物化學(xué)分析法[8-9]。
溶解氧是指溶解于水中分子狀態(tài)的氧,是水生物生存必不可少的條件。對于水產(chǎn)養(yǎng)殖業(yè)來說,水體溶解氧對水中生物的生存有著至關(guān)重要的影響,能夠反映出水體受到有機物污染的程度,它是水體污染程度的重要指標(biāo),也是衡量水質(zhì)的綜合指標(biāo)之一[10]。目前常用的溶解氧檢測方法有碘量法、電化學(xué)法(電流測定法、電導(dǎo)測定法)、熒光淬滅法等。
pH值作為水的最基本性質(zhì),它可以影響水體的弱酸、弱堿的離解程度,降低氯化物、氨、硫化氫等的毒性,對水質(zhì)的變化、生物繁殖的消長、腐蝕性、水處理效果等均有影響,是評價水質(zhì)的一個重要參數(shù)。pH值的傳統(tǒng)測量方式有化學(xué)分析法、試紙法和電位法等。
與水質(zhì)檢測的化學(xué)方法相比,基于光譜分析的水質(zhì)監(jiān)測技術(shù)是一個重要發(fā)展方向,已有工作表明,幾個重要水質(zhì)參數(shù)在光譜區(qū)均有很強的吸收。在一定的條件下,有機物的吸光度與有很好的相關(guān)性,利用這種相關(guān)性,可以用光譜技術(shù)直接測定[11-13]。
1.7.1 光譜法水質(zhì)監(jiān)測的理論基礎(chǔ)
光譜法則是基于朗伯比爾定律,通過監(jiān)測水產(chǎn)養(yǎng)殖水質(zhì)對特定波長的光的吸光度,然后對比存儲的標(biāo)準(zhǔn)曲線計算出水樣的值,屬于利用光譜學(xué)原理和試驗方法確定物質(zhì)結(jié)構(gòu)和化學(xué)成分的分析方法。通過建立有機物污染綜合指標(biāo)與水樣的光譜數(shù)據(jù)之間的回歸模型,來預(yù)測有機污染綜合指標(biāo)。
1.7.2 光譜法水質(zhì)方法步驟及試驗設(shè)備
COD在紫外254 nm處有很強的特征吸收相關(guān)性,利用這一選擇性吸收原理,可建立特定波長處吸光度值與COD濃度值的關(guān)系,計算溶液中COD濃度??偟庾V監(jiān)測方法有:堿性過硫酸鉀紫外分光光度法和氣相分子吸收光譜法。硝酸鹽是最穩(wěn)定的無機氮化合物,是亞硝酸鹽、氨氮和含氮有機物轉(zhuǎn)化的最終產(chǎn)物。目前的主要方法是堿性過硫酸鉀紫外分光光度法,該方法是采用堿性過硫酸鉀氧化,使有機氮和無機氮化合物轉(zhuǎn)變?yōu)橄跛猁}氮后紫外分光光度法進行測定[14]??偭自谥行詶l件下用過硫酸鉀(或硝酸-高氯酸)使試樣消解,對消解液用抗壞血酸溶液和鉬酸銨溶液處理,利用分光光度法進行測量。重金屬光譜監(jiān)測技術(shù)有原子吸收光譜法、分光光度法、熒光分析法,其中原子吸收光譜法,具有靈敏度高、檢出限低、分析速度快、選擇性好、抗干擾能力強等優(yōu)點,是目前測定重金屬含量最主要的方法。由于水產(chǎn)養(yǎng)殖水質(zhì)不同于廢水、地下水等,其對氧、氮、磷等元素特殊需求,使得其組成成分復(fù)雜,干擾監(jiān)測結(jié)果。針對易受干擾的指標(biāo),需對干擾物質(zhì)進行光譜分析,并與需監(jiān)測的物質(zhì)進行比較,確定利用光譜法測量水產(chǎn)養(yǎng)殖水質(zhì)該指標(biāo)的主要干擾物質(zhì),所遵從的原則是在不影響該指標(biāo)測量準(zhǔn)確度的前提下,盡可能減少干擾物質(zhì)種類?;诠庾V的水質(zhì)重要參數(shù)的監(jiān)測方法如表1所示,其中關(guān)于基于光譜技術(shù)的水質(zhì)COD監(jiān)測方面研究較多,技術(shù)較成熟、簡便;關(guān)于總磷、總氮、重金屬的檢測需要采用化學(xué)試劑進行預(yù)處理,操作有一定的復(fù)雜性。實現(xiàn)利用光譜技術(shù)對水產(chǎn)養(yǎng)殖水質(zhì)多參數(shù)的監(jiān)測,并提高光譜法水質(zhì)多參數(shù)監(jiān)測精度是值得探討的研究難點。
表1 基于光譜技術(shù)的水質(zhì)監(jiān)測方法
應(yīng)用光譜法進行水樣的定性或定量分析,提取待測水樣光譜信息需要進行光譜數(shù)據(jù)的處理,光譜數(shù)據(jù)處理分為預(yù)處理和光譜特征波段選擇2部分。
光譜中常常包含一些與待測樣品性質(zhì)無關(guān)聯(lián)的干擾信息,為了使建立的定性或定量分析模型更加穩(wěn)健、可靠,常常需要對測定的光譜數(shù)據(jù)進行預(yù)處理。常見的光譜特征波段選擇方法包括Savitzky-Golay平滑算法、小波分析、多元散射校正,3種常見的常見預(yù)算法的對比分析如表2所示。
2.1.1 Savitzky-Golay平滑算法
Savitzky-Golay算法是一種基本圖像處理方法,由Savitzky等在1964年首次提出[23],是一種在時域內(nèi)基于局域多項式最小二乘法擬合的濾波方法,通過卷積運算對曲線鄰域的像素灰度進行平均化,從而減少雜點、降低曲線對比度,該平滑算法做一種加權(quán)平均的過程。
表2 3種預(yù)處理算法的對比分析
SG平滑算法可用于對光譜數(shù)據(jù)作平滑處理[24],程長闊等[25]建立了紫外吸收光譜海水硝酸鹽反演模型,試驗結(jié)果顯示,SG卷積平滑能夠極大地降低模型預(yù)測誤差。李毛毛等[26-27]將SG平滑算法結(jié)合其它算法,以達到更好的去噪效果。喬星星等[28-29]對Savitzky- Golay平滑算法不同程度模式處理效果進行了研究,結(jié)果顯示,所建模型預(yù)測效果較未處理前有很大改善。Savitzky-Golay平滑算法不受樣本數(shù)據(jù)限制,適用于各種信號的平滑去噪,與傳統(tǒng)算法相比,該算法具有更穩(wěn)定、誤差更小的平滑去噪效果[30]。因此,SG濾波器通常用于光譜分析數(shù)據(jù)預(yù)處理,對原始數(shù)據(jù)進行平滑與去噪。
2.1.2 小波分析
小波分析是一種窗口大小固定但其形狀可變,時間窗和頻率窗都可以改變的時頻局部化分析方法[31-32]。該方法在低頻部分具有較高的頻率分辨率和較低的時間分辨率,在高頻部分具有較高的時間分辨率和較低的頻率分辨率,與傅里葉變換相比,小波變換是時間(空間)頻率的局部化分析,通過伸縮平移運算對光譜信息逐步進行多尺度細化,最終達到高頻處時間細分,低頻處頻率細分,能自動適應(yīng)時頻信號分析的要求,因此非常適合分析突變信息和非平穩(wěn)信息,把噪聲信息從正常信息中分離出來,達到去噪的目的[33]。
趙進輝等[34-37]采用小波分析法對農(nóng)產(chǎn)品的光譜數(shù)據(jù)進行去噪處理,相比于未去噪處理,均方根誤差明顯減小。Ma等[38-39]對小波分析方法進行了改進,并成功應(yīng)用于水樣的光譜數(shù)據(jù)的去噪處理。小波包去噪方法是小波分解的推廣,它提供了更豐富的信號分析方法[40]。張瑤等[41-42]利用小波包光譜信息進行去噪處理,結(jié)果表明,小波分析技術(shù)能夠有效地提高光譜預(yù)測效果。小波分析由于具有低熵性、多分辨率、去相關(guān)性和選基靈活性的特點, 能夠滿足各種去噪要求,廣泛應(yīng)用于去除光譜背景噪音、儀器干擾方面。
2.1.3 多元散射校正
多元散射校正(multiplicative scatter correction)最早是由Naes和Isaksson在1988年提出[43]。多元散射校正算法的基本思想是:假設(shè)每條光譜曲線都存在一條與其具有高相關(guān)性的理想光譜。真正理想的光譜雖然沒有辦法獲取,但通過使用樣本建模集的平均光譜曲線可以近似的替代,實現(xiàn)光譜數(shù)據(jù)的散射校正。多元散射校正的實現(xiàn)步驟如下:1)計算光譜平均值2)進行線性回歸運算,得出樣品的均勻程度、樣品特有的光譜信息3)通過樣品的均勻程度、樣品特有的光譜信息,進行光譜校正。
多元散射校正方法能夠剔除各樣品間由于散射影響所導(dǎo)致的基線變化影響[44-45],蘆永軍等[46]經(jīng)過試驗驗證得到的散射校正相關(guān)光譜有效地降低了散射的影響。湯斌等[47]運用多元散射校正法對受濁度影響的水樣光譜進行校正試驗,結(jié)果表明:該方法可在不影響水樣紫外-可見吸收光譜特征的前提下對其吸收曲線進行有效的校正。多元散射校正算法可提高原吸收光譜的信噪比,對消除光譜數(shù)據(jù)的線性散射干擾有較好的效果,該算法多用于光譜數(shù)據(jù)和濃度信息線性相關(guān)性較好的情況。
光譜儀獲取的光譜數(shù)據(jù)量大,光譜矩陣大量的冗余數(shù)據(jù),光譜矩陣中的無關(guān)信息等因素,導(dǎo)致光譜分析的速度變慢、效率降低。因此,從采集到的光譜數(shù)據(jù)中提取有益于建模的波長變量,去除冗余變量和無信息變量,可以提高光譜監(jiān)測的精度,優(yōu)化預(yù)測模型的性能。常見的光譜特征波段選擇方法包括連續(xù)投影算法、無信息變量消除、主成分分析法等[48]。3種常見的常見特征提取算法的對比分析如表3所示。
2.2.1 連續(xù)投影算法
連續(xù)投影算法(successive projections algorithm, SPA)是一種使矢量空間共線性最小化的前向變量選擇算法,其目標(biāo)是為了解決建模變量的共線性問題,改善多變量的建模預(yù)測效果。SPA算法的思想是:采用對光譜數(shù)據(jù)投影進行映射的方法構(gòu)造新的變量集,并對新的變量預(yù)測效果進行評價[49]。SPA算法的步驟:假設(shè)提取的特征波段的數(shù)量為,1)隨機選取光譜矩陣中的一列;2)計算該對剩余列的投影;3)重復(fù)第二步,直到得到個波段,停止迭代。
表3 3種常見特征提取算法的對比分析
周竹等[50-52]采用SPA算法對農(nóng)產(chǎn)品光譜數(shù)據(jù)進行特征波段的選擇,確定了最佳波長,降低了模型復(fù)雜度并提高了預(yù)測精度。國內(nèi)許多學(xué)者SPA光譜特征選擇算法進行了改進[53-56],郝勇等[57]引入蒙特卡羅方法,對SPA算法進行改進,對葡萄酒和蘋果的原始光譜進行酒精度和可溶性固形物信息的提取,解決了小樣本數(shù)據(jù)集變量選擇的問題。連續(xù)投影算法廣泛應(yīng)用于光譜領(lǐng)域,是一種最常用的光譜特征波段選擇的算法。
2.2.2 無信息變量去除算法
無信息變量消除算法(uninformative variables elimination,UVE)是在偏最小二乘回歸系數(shù)的基礎(chǔ)上建立的特征波段提取算法,用于去除對建立模型沒有貢獻的變量,即去除無信息變量[58]。UVE算法流程如下:1)把相同于自變量矩陣的變量數(shù)目的隨機變量矩陣(等同于噪音)加入光譜矩陣;2)通過交叉驗證的逐一剔除法建立PLS模型,得到回歸系數(shù)矩陣,分析回歸系數(shù)矩陣中回歸系數(shù)向量的平均值和標(biāo)準(zhǔn)偏差的商的穩(wěn)定性;3)根據(jù)該列光譜數(shù)據(jù)的商絕對值大小確定是否把改列變量用于PLS回歸模型中。
UVE算法能夠減少模型輸入變量的數(shù)量,降低建模的復(fù)雜性,廣泛用于光譜數(shù)據(jù)特征波段選擇[59]。Tan等[60]提出了基于無關(guān)信息變量消除多變量校正策略,經(jīng)驗證,該方法準(zhǔn)確性高、魯棒性強。Cai等[61]在光譜定量分析中,根據(jù)蒙特卡洛原理對無關(guān)信息算法進行優(yōu)化,消除穩(wěn)定差的變量的無關(guān)信息變量,該方法能夠光譜數(shù)據(jù)中選取重要波長,使預(yù)測結(jié)果更加可靠、準(zhǔn)確。Zhou等[62-63]將UVE與SPA結(jié)合,對光譜數(shù)據(jù)進行特征波段的選擇,發(fā)現(xiàn)與直接采用SPA算法相比,該算法參考更少的變量達到更高的預(yù)測效果。無信息變量去除算法能夠剔除沒有貢獻的變量,以達到光譜特征波段選擇的目的。
2.2.3 主成分分析
主成分分析(principal component analysis,PCA)將原變量通過線性組合變換為新變量,變換后的新變量相互正交、互不相關(guān),以排除信息中重疊的多余部分,并盡可能的保持原變量的數(shù)據(jù)信息。主成分分析法分析水樣紫外吸收光譜的基本思想是:將原來具有一定相關(guān)度的個波長的吸光度參數(shù),重新組合成一組較少個數(shù)的互不相關(guān)的吸收向量[64-69]。PCA算法的步驟:1)對光譜矩陣進行中心化2)計算光譜信息的協(xié)方差矩陣3)對矩陣進行特征值分解4)取出最大的個特征值對應(yīng)的特征向量,將所有的特征向量標(biāo)準(zhǔn)化后,組成特征向量矩陣5)對光譜矩陣中的每一個樣本,點乘特征向量矩陣,轉(zhuǎn)化為新的樣本。
主成分分析法可簡化水質(zhì)成分多樣性等問題,Assaad等[70-72]通過主成分分析,提取特征光譜數(shù)據(jù)解決水樣成分的多樣性和可變性等問題的影響。PCA算法也常與其它算法結(jié)合對水質(zhì)光譜信息進行簡化[73]。趙友全等[74]采用主成分分析結(jié)合歐氏距離和偏最小二乘法對水樣分類對COD含量的預(yù)測進行了定性和定量的分析。通過試驗驗證了該方法對實際水樣可以進行有效分類。主成分分析法是線性降維方法的基礎(chǔ),是一個典型的高維數(shù)據(jù)的降維方法,該方法最大優(yōu)勢在于可極大地縮短分類時間,常用于定性分析。
光譜建模常用的算法有偏最小二乘、最小二乘支持向量機和人工神經(jīng)網(wǎng)絡(luò)等。3種常見建模方法的對比分析如表4所示,其中偏最小二乘算法是線性建模算法,通常用于建立光譜數(shù)據(jù)和待測物質(zhì)之間具有線性相關(guān)的模型;而最小二乘支持向量機和人工神經(jīng)網(wǎng)絡(luò)算法是非線性建模算法,通常用于建立光譜數(shù)據(jù)和待測物質(zhì)之間具有非線性關(guān)系的預(yù)測模型。
表4 3種常見建模方法的對比分析
偏最小二乘法(partial least squares,PLS)最早于十六世紀(jì)晚期由H.Wold在計量經(jīng)濟學(xué)領(lǐng)域提出,是一種最常用的光譜建模方法,從廣義上講,相當(dāng)于主成分分析、多元線性回歸和典型相關(guān)分析的組合,其數(shù)學(xué)基礎(chǔ)為主成分分析,但它比主成分回歸更進了一步,主成分回歸只對自變量矩陣進行主成分分解,而偏最小二乘法將因變量矩陣和自變量矩陣同時進行主成分分解[72-73]。PLS算法步驟:1)同時對光譜數(shù)據(jù)矩陣和待測指標(biāo)矩陣進行因子分析,提取出相應(yīng)的隱含變量2)將隱含變量按照其對建模的貢獻率大小進行排序3)選擇最優(yōu)個數(shù)的隱含變量進行回歸。
針對水質(zhì)重要指標(biāo),PLS模型具有較好的效果[74-79]。Song等[80,81]采用建立GA-PLS校正數(shù)學(xué)模型,試驗表明,預(yù)測模型效果穩(wěn)健。楊鵬程等[79]利用紫外光譜技術(shù)結(jié)合偏最小二乘回歸(PLSR)方法,可很好地觀察長海水中硝酸鹽濃度的變化,對水質(zhì)進行監(jiān)測。Chen等[82-83]通過紫外可見光譜技術(shù)建立PLS模型,分別對水中COD和重金屬離子濃度進行分析監(jiān)測,結(jié)果顯示,預(yù)測值與真實值之間有極高的相關(guān)性。Dahlén等[84-86]采用PLS模型對COD、硝酸鹽等多個水質(zhì)指標(biāo)進行同時測定。PLS算法能顯著壓縮高維數(shù)據(jù),有效消除變量之間的多重共線性,充分提取因變量矩陣與自變量矩陣中的有效信息,通過減少光譜數(shù)據(jù)計算量來提高模型性能,利用該算法可建立簡便的光譜預(yù)測模型。
支持向量機(support vector machine,SVM)是二十世紀(jì)九十年代興起的一種機器學(xué)習(xí)方法[87],它遵循結(jié)構(gòu)風(fēng)險最小化原則,能解決傳統(tǒng)機器學(xué)習(xí)中在小樣本、非線性等情形下常見的陷入局部最優(yōu)以及過學(xué)習(xí)等問題,對于非線性建模、解決樣本量偏少和數(shù)據(jù)挖掘領(lǐng)域具有很強的能力。支持向量機思想:1)線性可分情況,把問題轉(zhuǎn)化為一個凸優(yōu)化問題,用拉格朗日乘子法簡化,然后用既有的算法解決;2)線性不可分,用核函數(shù)將樣本投射到高維空間,使其變成線性可分的情形,利用核函數(shù)來減少高緯度計算量。最小二乘支持向量機(least squares support vector machine ,LS-SVM)是一種經(jīng)過改進的支持向量機方法,將其約束條件由不等式改為等式,轉(zhuǎn)換為在對偶空間中對一個等式方程組進行二次規(guī)劃問題的求解,在高維空間里求解最小化損失函數(shù)[88]。
Choi等[89-91]建立了最小二乘支持向量機法對水質(zhì)進行預(yù)測。國內(nèi)外許多學(xué)者對LS-SVM模型進行改進,以提高水質(zhì)預(yù)測模型的性能。曹泓等[92-93]對紫外、紅外多源光譜特征組合建立LS-SVM模型,對化學(xué)需氧量進行定量預(yù)測,良好的預(yù)測精度。最小二乘支持向量機可以極大的提高模型的計算效率,可以發(fā)揮小樣本、泛化能力強等優(yōu)點,在保證預(yù)測準(zhǔn)確的同時,縮短了光譜分析預(yù)測模型的運行時間。
人工神經(jīng)網(wǎng)絡(luò)(artificial neural network,ANN)是在現(xiàn)代神經(jīng)科學(xué)研究成果的基礎(chǔ)上提出的,是應(yīng)用類似于大腦神經(jīng)突觸聯(lián)接的結(jié)構(gòu)進行信息處理的數(shù)學(xué)模型。該算法可以在輸入變量和輸出變量之間建立高度非線性的映射模型,在映射過程中能夠并行分布處理和自適應(yīng)學(xué)習(xí)。人工神經(jīng)網(wǎng)絡(luò)的種類有很多,包括感知器人工神經(jīng)網(wǎng)絡(luò)、反向傳播人工神經(jīng)網(wǎng)絡(luò)、人工神經(jīng)網(wǎng)絡(luò)和自組織人工神經(jīng)網(wǎng)絡(luò)等,目前在光譜分析和建模中得到廣泛的應(yīng)用。BP神經(jīng)網(wǎng)絡(luò)通常由一個3層網(wǎng)絡(luò)組成,分別稱為輸出層、隱含層和輸入層。BPNN的輸入層、輸出層和隱含層都是由神經(jīng)元構(gòu)成。信號從輸入層神經(jīng)元輸入后,傳至隱含層神經(jīng)元,經(jīng)過隱含層傳遞函數(shù)計算之后,將輸出的信號傳遞到輸出層,最終由輸出層得到模型的計算結(jié)果[94]。在建立BPNN模型的過程中,通過將樣本已知的結(jié)果和模型的輸出結(jié)果進行對比,如果輸出結(jié)果的預(yù)測誤差沒有滿足設(shè)定的要求,則通過反復(fù)迭代的方法,直到限定的迭代次數(shù)達到或者預(yù)測均方根誤差小于設(shè)定的閾值。
Zakaluk等[95-96]利用人工神經(jīng)網(wǎng)絡(luò)算法來提高水質(zhì)預(yù)測精度的方法。分析對比多種人工神經(jīng)網(wǎng)絡(luò)模型,發(fā)現(xiàn)徑向基函數(shù)神經(jīng)網(wǎng)絡(luò)(RBFNN)和反向傳播人工神經(jīng)網(wǎng)絡(luò)(BP人工神經(jīng)網(wǎng)絡(luò))對水產(chǎn)養(yǎng)殖水質(zhì)預(yù)測效果更突出[97]。徑向基函數(shù)神經(jīng)網(wǎng)絡(luò)模型常用于水質(zhì)的定量與定性分析,Xie等[98]將NIR技術(shù)應(yīng)用于水摻入的楊梅汁的監(jiān)測,采用最優(yōu)參數(shù)的RBFNN模型可分離純種楊梅汁樣品。Mesquita等[99-100]提出了一種紫外多波長與BP神經(jīng)網(wǎng)絡(luò)相結(jié)合的有機廢水COD預(yù)測技術(shù),誤差分析數(shù)據(jù)顯示相對誤差控制在5%以內(nèi)。BP人工神經(jīng)網(wǎng)絡(luò)是目前應(yīng)用最廣泛的人工神經(jīng)網(wǎng)絡(luò)算法,Ji等[101]建立了一種利用BP神經(jīng)網(wǎng)絡(luò)和動力學(xué)分光光度法同時測定自來水中鐵和鎂的分析方法。BP神經(jīng)網(wǎng)絡(luò)模型常用于對在水產(chǎn)養(yǎng)殖水質(zhì)的監(jiān)測與預(yù)警[102-103],Qu等[104]開發(fā)一種可見的近紅外成像技術(shù),建立了BP人工神經(jīng)網(wǎng)絡(luò)模型,結(jié)果證明,該模型可快速預(yù)測在水產(chǎn)養(yǎng)殖環(huán)境中腐植酸鈉的含量,進一步實時監(jiān)控水產(chǎn)養(yǎng)殖水的質(zhì)量。人工神經(jīng)網(wǎng)絡(luò)有自學(xué)習(xí)、高容錯和高度非線性描述能、高速尋找優(yōu)化解的能力等優(yōu)點,避免了光譜分析模型計算量大、計算速度慢等問題。目前,相較于其它人工神經(jīng)網(wǎng)絡(luò),BP神經(jīng)網(wǎng)絡(luò)是應(yīng)用最廣泛的水質(zhì)預(yù)測建模方法。
基于光譜技術(shù)水質(zhì)預(yù)測模型對比如表5所示,通過分析可知,對于小樣本數(shù)據(jù)偏最小二乘算法預(yù)測效果最好,偏最小二乘支持向量機算法經(jīng)過改進,預(yù)測效果明顯增強。
表5 基于光譜技術(shù)水質(zhì)預(yù)測模型對比
注:PCA-PSO-ELM(principal component analysis-particle swarm optimization-extreme learning machine)是基于主成分分析聯(lián)合粒子群優(yōu)化極限學(xué)習(xí)機預(yù)測模型,PCA-PSO-LS-SVM(principal component analysis-particle swarm optimization-least squares support vector machine)是基于主成分分析聯(lián)合粒子群優(yōu)化最小二乘支持向量機預(yù)測模型,NMF-PSO-LS-SVM(non-negative matrix factorization- particle swarm optimization-least squares support vector machine)是基于非負矩陣分解聯(lián)合粒子群優(yōu)化最小支持向量機預(yù)測模型,RMSEP(root-mean-square error of prediction)是預(yù)測誤差均方根
基于光譜技術(shù)的水質(zhì)監(jiān)測突破了傳統(tǒng)檢測方法的操作復(fù)雜、不可重復(fù)、易造成附加污染等局限,成為了水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測的重要方法。
1)目前,隨著食品質(zhì)量安全問題的日益突出以及水產(chǎn)養(yǎng)殖水質(zhì)污染頻繁發(fā)生,迫切地需要構(gòu)建一種在線、實時的水質(zhì)監(jiān)測系統(tǒng),實現(xiàn)對水質(zhì)異常狀況進行預(yù)警?,F(xiàn)階段的水質(zhì)檢測往往需要結(jié)合一些實驗室處理方法,如化學(xué)分析法等,在做檢測結(jié)果之前,已經(jīng)消耗了一定的時間,因此水質(zhì)檢測無法做到實時在線進行。將光譜技術(shù)與實時在線監(jiān)測技術(shù)相結(jié)合,實現(xiàn)對水產(chǎn)養(yǎng)殖水質(zhì)進行實時在線監(jiān)測和預(yù)警,將對水質(zhì)監(jiān)測領(lǐng)域具有更大的實際意義。
2)多源光譜融合的水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測將會成為新的發(fā)展方向?,F(xiàn)階段的水質(zhì)監(jiān)測多采用單一光譜,無法達到較高的監(jiān)測精度。而將信息融合技術(shù)應(yīng)用于光譜領(lǐng)域,融合存在一定的相關(guān)性和互補性的不同光譜,可提高預(yù)測模型的分析精度和魯棒性。
3)利用光譜技術(shù)對水質(zhì)多參數(shù)監(jiān)測,是今后水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測研究的發(fā)展方向。由于水中懸浮物對不同波長可見光的散射存在非線性關(guān)系,且水中懸浮物對影響水質(zhì)參數(shù)的部分有機物存在吸附,導(dǎo)致單一可見光波長的濁度補償方法無法準(zhǔn)確地扣除濁度引起的散射干擾。因此,研究一種抵消濁度干擾,對測量光譜進行有效地校正的方法成為水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測的關(guān)鍵技術(shù)問題。
4)對于光譜數(shù)據(jù)的處理,將多種數(shù)據(jù)處理算法相結(jié)合,仍將占據(jù)主導(dǎo)。目前常見的數(shù)據(jù)處理方法是以2種或2種以上的算法融合的數(shù)據(jù)處理方法為主,在今后較長一段時間內(nèi),這種方法仍會占據(jù)主導(dǎo)。常見的如蒙特卡羅方法結(jié)合連續(xù)投影算法(CARS-SPA)預(yù)處理算法,無信息變量消除算法結(jié)合連續(xù)投影算法(UVE-SPA)特征波段提取算法,等。將多種數(shù)據(jù)處理算法相結(jié)合,對傳統(tǒng)算法進行改進,能夠更好地發(fā)揮這些算法的優(yōu)勢,以實現(xiàn)精確、快速地提取水質(zhì)參數(shù)有效的光譜信息。
5)非線性數(shù)據(jù)建模,將成為光譜技術(shù)應(yīng)用于水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測的主流建模發(fā)方法。水環(huán)境是一個無序的、非穩(wěn)定的、非平衡的隨機系統(tǒng),不同元素之間往往存在著隨機性、協(xié)同現(xiàn)象和相干效應(yīng),非線性建模算法可增加監(jiān)測的準(zhǔn)確性、快速性、魯棒性。
[1] Pu H, Liu D, Qu J H, et al. Applications of imaging spectrometry in inland water quality monitoring: A review of recent developments[J]. Water Air & Soil Pollution, 2017, 228(4): 131.
[2] Rojas F S. Process analytical chemistry: Applications of ultraviolet/visible spectrometry in environmental analysis: An overview[J]. Applied Spectroscopy Reviews, 2009, 44(3): 245-265.
[3] 李進,張葆宗. 反滲透水處理系統(tǒng)微生物污染特性分析及對策[J]. 工業(yè)水處理,2000,20(5):10-12. Li Jin, Zhang Baozong. Pollution and prevention of microorganism on reverse osmosis film[J]. Industrial Water Treatment, 2000, 20(5): 10-12. (in Chinese with English abstract)
[4] 李玉春. 基于紫外可見光譜的水下多參數(shù)水質(zhì)檢測技術(shù)研究[D]. 天津:天津大學(xué),2012.
[5] 侯迪波,張堅,陳泠,等. 基于紫外-可見光光譜的水質(zhì)分析方法研究進展與應(yīng)用[J]. 光譜學(xué)與光譜分析,2013,33(7):1839-1844. Hou Dibo, Zhang Jian, Chen Ling, et al. Water quality analysis by UV-Vis spectroscopy: A review of methodology and application[J]. Spectroscopy and Spectral Analysis, 2013, 33(7): 1839-1844. (in Chinese with English abstract)
[6] Cuesta A, Todoli J L, Canals A. Flow injection method for the rapid determination of chemical oxygen demand based on microwave digestion and chromium speciation in flame atomic absorption spectrometry[J]. Journal of SpectrochimicaActa. Part B-Atomic Spectroscopy, 1996, 51(14): 1791-1800.
[7] 唐慧穎. 連續(xù)流動分析法測定水中的總磷[J]. 污染防治技術(shù), 2014(1): 51-54.
[8] Wei F X, Ma X Z, Lei L G, et al. Detection of total nitrogen in water with microwave digestion cadmium column and spectrophotometry[J]. Journal of Analytical Science, 2011, 27(5): 615-618.
[9] Louren?o N D, Lopes J A, Almeida C F, et al. Bioreactor monitoring with spectroscopy and chemometrics: A review[J]. Analytical & Bioanalytical Chemistry, 2012, 404(4): 1211-1237.
[10] Yang Y, Yan G, Lin Q. Determination of heavy metal ions in Chinese herbal medicine by microwave digestion and RP-HPLC with UV-Vis detection[J]. Microchimica Acta, 2004, 144(4): 297-302.
[11] Forzani E S, Zhang H, Chen W, et al. Detection of heavy metal ions in drinking water using a high-resolution differential surface plasmon resonance sensor[J]. Environmental Science and Technology, 2005, 39(5): 1257-1262.
[12] 周娜. BP人工神經(jīng)網(wǎng)絡(luò)紫外吸收光譜法直接測定COD研究[D]. 成都:四川大學(xué),2006.
[13] 國家環(huán)??偩郑蛷U水監(jiān)測分析方法編委會編. 水和廢水監(jiān)測分析方法[M]. 第四版. 北京:中國環(huán)境科學(xué)出版社,2002.
[14] 張國強. UV法測量水質(zhì)COD技術(shù)研究[D]. 成都:電子科技大學(xué),2007.
[15] 郝瑞霞,曹可心,趙鋼,等. 用紫外光譜參數(shù)表征污水中溶解性有機污染物[J]. 北京工業(yè)大學(xué)學(xué)報,2006,32(12):1062-1066. Hao Ruixia, Cao Kexin, Zhao Gang, et al. Ultraviolet absorption spectrum characterization approach for quantitative analysis of dissolved organic contaminants in sewage[J]. Journal of Beijing University of Technology, 2006, 32(12): 1062-1066. (in Chinese with English abstract)
[16] Langergraber G, Fleischmann N, Hofst?dter F. A multivariate calibration procedure for UV/VIS spectrometric quantification of organic matter and nitrate in wastewater[J]. Water Science and Technology, 2003, 47(2): 63-71.
[17] 王睿,余震虹,魚瑛. 紫外吸收光譜法研究硝酸鹽溶液[J]. 光譜實驗室,2009,26(2):206-209. Wang Rui, Yu Zhenhong, Yu Ying. Investigation on nitrate solution by ultraviolet absorption spectrometry[J]. Chinese Journal of Spectroscopy Laboratory, 2009, 26(2): 206-209. (in Chinese with English abstract)
[18] 海彩虹. 紫外分光光度法測定濃維磷糖漿中的總磷量[J]. 中國藥業(yè),2007,16(19):22-23.
[19] 王斌,楊慧中. 一種水質(zhì)總磷在線檢測的光譜數(shù)據(jù)處理方法[J]. 激光與光電子學(xué)進展,2015,52(4):236-241.
[20] 肖錫林,魏永卷,薛金花,等. Pb-XO配合物顯色光度法測定水樣中微量鉛[J]. 應(yīng)用化工,2009,38(2):296-299. Xiao Xilin, Wei Yongjuan, Xue Jinhua, et al. Spectrophotometric determination of trace lead in water samples with Pb(Ⅱ)-XO system[J]. Applied Chemical Industry, 2009, 38(2): 296-299. (in Chinese with English abstract)
[21] 王娟,張飛,王小平,等. 平行因子法結(jié)合自組織映射神經(jīng)網(wǎng)絡(luò)的三維熒光特征及其與水質(zhì)的關(guān)系[J]. 光學(xué)學(xué)報,2017,37(7):349-359.
[22] 杜艷紅,張偉玉,楊仁杰,等. 基于可見-近紅外光譜的水質(zhì)pH值分析[J]. 湖北農(nóng)業(yè)科學(xué),2012,51(3):612-614.
[23] Savitzky A, Golay M J E. Smoothing+differentiation of data by simplified least squares procedures[J]. Analytical Chemistry, 1964, 36(8): 1627-1630.
[24] 潘磊慶,劉明,韓東海,等. 水蜜桃貨架期內(nèi)糖度的近紅外光譜檢測[J]. 南京農(nóng)業(yè)大學(xué)學(xué)報,2013,36(4):116-120. Pan Leiqing, Liu Ming, Han Donghai, et al. Detection of the sugar content of juicy peach during shelf life by near infrared spectroscopy technology[J]. Journal of Nanjing Agricultural University, 2013, 36(4): 116-120. (in Chinese with English abstract)
[25] 程長闊. 紫外吸收光譜法海水硝酸鹽測量系統(tǒng)設(shè)計與研究[D]. 大連:大連海洋大學(xué),2015.
[26] 李毛毛,鄭喜群,任健,等. 近紅外光譜法快速檢測甜菜糖度的模型優(yōu)化[J]. 食品安全質(zhì)量檢測學(xué)報,2015(8):3026-3029. Li Maomao, Zheng Xiqian, Ren Jian, et al. Model optimization on rapid detection of beet sugar content by near infrared spectroscopy[J]. Journal of Food Safety and Quality, 2015(8): 3026-3029. (in Chinese with English abstract)
[27] 文韜,鄭立章,龔中良,等. 基于近紅外光譜技術(shù)的茶油原產(chǎn)地快速鑒別[J]. 農(nóng)業(yè)工程學(xué)報,2016,32(16): 293-299. Wen Tao, Zheng Lizhang, Gong Zhongliang, et al. Rapid identification of geographical origin of camellia oil based on near infrared spectroscopy technology[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2016, 32(16): 293-299. (in Chinese with English abstract)
[28] 喬星星,馮美臣,楊武德,等. SG平滑處理對冬小麥地上干生物量光譜監(jiān)測的影響[J]. 山西農(nóng)業(yè)科學(xué),2016,44(10):1450-1454. Qiao Xingxing, Feng Meichen, Yang Wude, et al. Effect of SG smoothing processing on predicting the above ground dry biomass of winter wheat[J]. Journal of Shanxi Agricultural Sciences, 2016, 44(10): 1450-1454. (in Chinese with English abstract)
[29] 陳華舟,潘濤,陳潔梅. 多元散射校正與Savitzky-Golay平滑模式的組合優(yōu)選應(yīng)用于土壤有機質(zhì)的近紅外光譜分析[J]. 計算機與應(yīng)用化學(xué),2011,28(5):518-522. Chen Huazhou, Pan Tao, Chen Jiemei. Combination optimization of multiple scatter correction and Savitzky- Golay smoothing models applied to the near infrared spectroscopy analysis of soil organic matter[J]. Computers and Applied Chemistry, 2011, 28(5): 518-522. (in Chinese with English abstract)
[30] 雷林平. 基于Savitzky-Golay算法的曲線平滑去噪[J]. 電腦與信息技術(shù),2014,22(5):30-31. Lei Linping. Curve smooth denoising based on Savitzky- Golay algorithm[J]. Computer and Information Technology, 2014, 22(5): 30-31. (in Chinese with English abstract)
[31] 付小葉. 傅里葉變換與小波分析[J]. 數(shù)學(xué)建模及其應(yīng)用,2016,5(2):83-84.
[32] 文莉,劉正士,葛運建. 小波去噪的幾種方法[J]. 合肥工業(yè)大學(xué)學(xué)報,2002,25(2):167-172. Wen Li, Liu Zhengshi, Ge Yunjian. Several methods of wavelet denoising[J]. Journal of Hefei University of Technology, 2002, 25(2): 167-172. (in Chinese with English abstract)
[33] 苑津莎,張冬雪,李中. 基于改進閾值法的小波去噪算法研究[J]. 華北電力大學(xué)學(xué)報(自然科學(xué)版),2010,37(5):92-97. Yuan Jinsha, Zhang Dongxue, Li Zhong. Wavelet denoising algorithm based on improved thresholding method[J]. Journal of North China Electric Power University, 2010, 37(5): 92-97. (in Chinese with English abstract)
[34] 趙進輝,袁海超,劉木華,等. 導(dǎo)數(shù)同步熒光光譜-小波- SGA-LSSVR聯(lián)用快速測定鴨蛋蛋清中新霉素殘留含量[J]. 分析化學(xué),2013,41(4):546-552. Zhao Jinhui, Yuan Haichao, Liu Muhua, et al. Rapid determination of neomycin content in duck egg white using derivative synchronous fluorescence-wavelet-subsection genetic algorithm-least squares support vector regression[J]. Chinese Journal of Analytical Chemistry, 2013, 41(4): 546-552. (in Chinese with English abstract)
[35] 羅霞,洪添勝,羅闊,等. 小波變換和連續(xù)投影算法在火龍果總酸無損檢測中的應(yīng)用[J]. 光譜學(xué)與光譜分析,2016,36(5):1345-1351. Luo Xia, Hong Tiansheng, Luo Kuo, et al. Application of wavelet transform and successive projections algorithm in the non-destructive measurement of total acid content of pitaya[J]. Spectroscopy and Spectral Analysis, 2016, 36(5): 1345-1351. (in Chinese with English abstract)
[36] 黃雙萍,岳學(xué)軍,洪添勝,等. 基于小波變換與LS-SVR的柑橘葉片磷含量高光譜監(jiān)測模型[J]. 廣東農(nóng)業(yè)科學(xué),2013,13:37-40. Huang Shuangping, Yue Xuejun, Hong Tiansheng, et al. Hyperspectrum based models for monitoring phosphorus content of Luogang Orange leaf using wavelet denoising and least squares support vector regression analysis[J]. Guangdong Agricultural Sciences, 2013, 40(13): 37-40. (in Chinese with English abstract)
[37] 陸宇振,周健民,余常兵,等. 應(yīng)用小波分析進行油菜籽紅外光聲光譜去噪[J]. 光譜實驗室,2013,30(5): 2126-2131. Lu Yuzhen, Zhou Jianmin, Yu Changbing, et al. Denoising of infrared photoacoustic spectra of rapeseeds using wavelet analysis[J]. Chinese Journal of Spectroscopy Laboratory, 2013, 30(5): 2126-2131. (in Chinese with English abstract)
[38] Ma Y, Zhang J, An N. Spectral fidelity analysis of compressed sensing reconstruction hyperspectral remote sensing image based on wavelet transformation[C]// Chinese Conference on Pattern Recognition. Springer, Berlin, Heidelberg, 2014: 138-148.
[39] 潘國鋒. 水質(zhì)總氮光譜檢測建模方法研究[D]. 無錫:江南大學(xué),2014.
[40] 譚孝賢. 支持向量機在小波包去噪方法中的應(yīng)用[D]. 上海:上海交通大學(xué),2009.
[41] 張瑤,鄭立華,李民贊,等. 基于光譜學(xué)原理與小波包分解技術(shù)預(yù)測蘋果樹葉片氮素含量[J]. 農(nóng)業(yè)工程學(xué)報,2013,29(增刊1):101-108. Zhang Yao, Zheng Lihua, Li Minzan, et al. Predicting apple tree leaf nitrogen content based on hyperspectral applying wavelet and wavelet packet analysis [J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2013,29(Supp.1): 101-108. (in Chinese with English abstract)
[42] 湯斌,魏彪,毛本將,等. 紫外-可見吸收光譜法水質(zhì)檢測系統(tǒng)的噪聲分析與處理研究[J]. 激光與光電子學(xué)進展,2014,51(4):201-207. Tang Bin, Wei Biao, Mao Benjiang, et al. Noise analysis and denoising research on the UV-Visible absorption spectroscopy water quality detection system[J]. Laser and Optoelectronics Progress, 2014,51(4): 201-207. (in Chinese with English abstract)
[43] Geladi P, MacDougall D, Martens H. Linearization and scatter-correction for near-infrared reflectance spectra of Meat[J]. Applied Spectroscopy, 1985, 39(3): 491-500.
[44] 吳德操,魏彪,熊雙飛,等. 針對水質(zhì)監(jiān)測的紫外-可見光譜雙光程融合優(yōu)化算法[J]. 光譜學(xué)與光譜分析,2017,37(3):799-805. Wu Decao, Wei Biao, Xiong Shuangfei, et al. An optimized ultraviolet-visible spectrum dual optical path length fusion algorithm for water quality monitoring [J]. Spectroscopy and Spectral Analysis, 2017, 37(3): 799-805. (in Chinese with English abstract)
[45] 王動民,紀(jì)俊敏,高洪智. 多元散射校正預(yù)處理波段對近紅外光譜定標(biāo)模型的影響[J]. 光譜學(xué)與光譜分析,2014,34(9):2387-2390.Wang Dongmin, Ji Junmin, Gao Hongzhi. The effect of MSC spectral pretreatment regions on near infrared spectroscopy calibration results[J]. Spectroscopy and Spectral Analysis, 2014, 34(9): 2387-2390. (in Chinese with English abstract)
[46] 蘆永軍,曲艷玲,宋敏. 近紅外相關(guān)光譜的多元散射校正處理研究[J]. 光譜學(xué)與光譜分析,2007,27(5):877-880. Lu Yongjun, Qu Yanling, Song Min. Research on the correlation chart of near infrared spectra by using multiple scatter correction technique[J]. Spectroscopy and Spectral Analysis, 2007, 27(5): 877-880. (in Chinese with English abstract)
[47] 湯斌,魏彪,吳德操,等. 一種紫外-可見光譜法檢測水質(zhì)COD的濁度影響實驗研究[J]. 光譜學(xué)與光譜分析,2014,34(11):3020-3024. Tang Bin, Wei Biao, Wu Decao, et al. Experimental research of turbidity influence on water quality monitoring of COD in UV-Visible spectroscopy[J]. Spectroscopy and Spectral Analysis, 2014, 34(11): 3020-3024. (in Chinese with English abstract)
[48] 王鳳花,朱海龍,戈振揚. 近紅外光譜數(shù)據(jù)建模方法的研究進展[J]. 農(nóng)業(yè)工程,2011,1(1):56-61. Wang Fenghua, Zhu Hailong, Ge Zhenyang. Progress of near-infrared spectral data modeling method[J]. Agricultural Engineering, 2011, 1(1): 56-61. (in Chinese with English abstract)
[49] Hurt N E. Signal enhancement and the method of successive projections[J]. Acta Applicandae Mathematica, 1991, 23(2): 145-162.
[50] 周竹,尹建新,周素茵,等. 基于近紅外光譜與連續(xù)投影算法的針葉材表面節(jié)子缺陷識別[J]. 激光與光電子學(xué)進展,2017,54(2):311-319. Zhou Zhu, Yin Jianxin, Zhou Suyin, et al. Knot defection on coniferous wood surface by near infrared spectroscopy and successive projections algorithm[J]. Laser and Optoelectronics Progress, 2017, 54(2): 311-319. (in Chinese with English abstract)
[51] 劉思伽,田有文,張芳,等. 采用二次連續(xù)投影法和BP人工神經(jīng)網(wǎng)絡(luò)的寒富蘋果病害高光譜圖像無損檢測[J]. 食品科學(xué),2017,38(8):277-282.
[52] Wang L, Pu H, Sun D W. Estimation of chlorophyll-a concentration of different seasons in outdoor ponds using hyperspectral imaging[J]. Talanta, 2016, 147: 422-429. (in Chinese with English abstract)
[53] 孫旭東,郝勇,蔡麗君,等. 基于抽取和連續(xù)投影算法的可見近紅外光譜變量篩選[J]. 光譜學(xué)與光譜分析,2011,31(9):2399-2402. Sun Xudong, Hao Yong, Cai Lijun, et al. Selection of visible-NIR variables based on extraction and successive projections algorithm[J]. Spectroscopy and Spectral Analysis, 2011, 31(9): 2399-2402. (in Chinese with English abstract)
[54] 劉明博,唐延林,李曉利,等. 水稻葉片氮含量光譜監(jiān)測中使用連續(xù)投影算法的可行性[J]. 紅外與激光工程,2014,43(4):1265-1271. Liu Mingbo, Tang Yanlin, Li Xiaoli, et al. Feasibility of using successive projections algorithm in spectral monitoring of rice leaves nitrogen contents[J]. Infrared and Laser Engineering, 2014, 43(4): 1265-1271. (in Chinese with English abstract)
[55] 姜微,房俊龍,王樹文,等. CARS-SPA算法結(jié)合高光譜檢測馬鈴薯還原糖含量[J]. 東北農(nóng)業(yè)大學(xué)學(xué)報,2016,47(2):88-95. Jiang Wei, Fang Junlong, Wang Shuwen, et al. Using CARS-SPA algorithm combined with hyperspectral to determine reducing sugars content in potatoes[J]. Journal of Northeast Agricultural University, 2016, 47(2): 88-95. (in Chinese with English abstract)
[56] 錢海波,孫來軍,王樂凱,等. 基于連續(xù)投影算法的小麥濕面筋近紅外校正模型優(yōu)化[J]. 中國農(nóng)學(xué)通報,2011,27(18):51-56. Qian Haibo, Sun Laijun, Wang Leikai, et al. Near infrared spectroscopy calibration model optimizing of wet gluten based on successive projections algorithm[J]. Chinese Agricultural Science Bulletin, 2011, 27(18): 51-56. (in Chinese with English abstract)
[57] 郝勇,孫旭東,王豪. 基于改進連續(xù)投影算法的光譜定量模型優(yōu)化[J]. 江蘇大學(xué)學(xué)報(自然科學(xué)版),2013,34(1):49-53. Hao Yong, Sun Xudong, Wang Hao. Spectral quantitative model optimization by modified successive projection algorithm[J]. Journal of Jiangsu University, 2013, 34(1): 49-53. (in Chinese with English abstract)
[58] Shao X, Wang F, Chen D, et al. A method for near-infrared spectral calibration of complex plant samples with wavelet transform and elimination of uninformative variables[J]. Analytical and Bioanalytical Chemistry, 2004, 378(5): 1382.
[59] 侯靜. 無信息變量消去法結(jié)合直接正交法用于近紅外光譜分析研究[D]. 哈爾濱:哈爾濱理工大學(xué),2013.
[60] Tan C, Wu T, Xu Z, et al. A simple ensemble strategy of uninformative variable elimination and partial least-squares for near-infrared spectroscopic calibration of pharmaceutical products[J]. Vibrational Spectroscopy, 2012, 58(1): 44-49. (in Chinese with English abstract)
[61] Cai W, Li Y, Shao X. A variable selection method based on uninformative variable elimination for multivariate calibration of near-infrared spectra[J]. Chemometrics and Intelligent Laboratory Systems, 2008, 90(2): 188-194.
[62] Zhou L, Ma W, Zhang H, et al. Developing a PCA–ANN model for predicting chlorophyll a, concentration from field hyperspectral measurements in Dianshan lake, China[J]. Water Quality Exposure & Health, 2015, 7(4): 1-12.
[63] Ye S, Wang D, Min S. Successive projections algorithm combined with uninformative variable elimination for spectral variable selection[J]. Chemometrics and Intelligent Laboratory Systems, 2008, 91(2): 194-199.
[64] Pearson K. Onlines and planes of closer fit to systems of points in space[J]. Philos.Mag,1901(2): 559-572.
[65] Hotelling H. Analysis of a complex of statistical variables into principal components[J]. Edu Psych,1933(24): 417-441.
[66] Anokhin V N, Batrakov G F, Zemlyanov A D, et al. X-Ray fluorescence analysis of the principal elements of the seawater salt composition in the tropical Atlantic[J]. Soviet Journal of Physical Oceanography, 1992, 3(3): 209-214.
[67] 張崢,魏彪,湯戈,等. 一種紫外-可見光譜法水質(zhì)COD檢測的預(yù)測模型研究[J]. 激光雜志,2016,37(4):21-24.
[68] 湯斌,魏彪,吳德操,等. 主元分析降維聯(lián)合廣義判別分類的水質(zhì)檢測紫外-可見光譜數(shù)據(jù)處理方法[J]. 激光雜志,2014(10):112-114.
[69] 唐紅,鄭文斌,李憲霞. 主成分分析在光全散射特征波長選擇中的應(yīng)用[J]. 光學(xué)精密工程,2010,18(8):1691-1698. Tang Hong, Zheng Wenbin, Li Xianxia. Application of principal component analysis to selection of characteristic wavelengths with total light scattering[J]. Optics and Precision Engineering, 2010, 18(8): 1691-1698. (in Chinese with English abstract)
[70] Assaad A, Pontvianne S, Pons M N. Assessment of organic pollution of an industrial river by synchronous fluorescence and UV-Vis spectroscopy: The Fensch River (NE France) [J]. Environmental Monitoring & Assessment, 2017, 189(5): 229.
[71] 劉雙印,徐龍琴,李振波,等. 基于PCA-MCAFA-LSSVM的養(yǎng)殖水質(zhì)pH值預(yù)測模型[J]. 農(nóng)業(yè)機械學(xué)報,2014,45(5):239-246. Liu Shuangyin, Xu Longqin, Li Zhengbo, et al. Forecasting model for pH value of aquaculture water quality based on PCA-MCAFA-LSSVM[J]. Transactions of the Chinese Society for Agricultural Machinery, 2014, 45(5): 239-246. (in Chinese with English abstract)
[72] Kong X, Liu Y, Jian H, et al. New approach for rapid assessment of trophic status of Yellow Sea and East China Sea using easy-to-measure parameters[J]. Journal of Ocean University of China, 2017, 16(5): 781-792.
[73] 買巍,趙曉明,張健飛,等. 紫外可見光譜結(jié)合多元分析法在線檢測混合染料溶液化學(xué)需氧量[J]. 光譜學(xué)與光譜分析,2017,37(7):2105-2109. Mai Wei, Zhao Xiaoming, Zhang Jianfei, et al. Multivariate calibration of a UV-Vis spectrophotometer used for online measurements of chemical oxygen demand in dyeing wastewater[J]. Spectroscopy and Spectral Analysis, 2017, 37(7): 2105-2109. (in Chinese with English abstract)
[74] 趙友全,李霞,劉瀟,等. 基于PCA的水質(zhì)紫外吸收光譜分析模型研究[J]. 光譜學(xué)與光譜分析,2016,36(11):3592-3596.
[75] Abdi H, Williams L J. Partial Least Squares Methods: Partial Least Squares Correlation and Partial Least Square Regression[J]. Methods in Molecular Biology, 2013, 930: 549-579.
[76] Fujiwara K, Sawada H, Kano M. Input variable selection for PLS modeling using nearest correlation spectral clustering [J]. Chemometrics and Intelligent Laboratory Systems, 2012, 118: 109-119.
[77] 劉飛. 水體COD的光譜學(xué)測量方法研究與傳感設(shè)備研制[D]. 重慶:重慶郵電大學(xué),2016.
[78] 張榮標(biāo),馮俊,謝志超. 基于廣義回歸神經(jīng)網(wǎng)絡(luò)的COD在線檢測方法研究[J]. 儀器儀表學(xué)報,2008,29(11): 2357-2361. Zhang Rongbiao, Feng Jun, Xie Zhichao. Study on COD on-line detection method based on general regression neural network[J]. Chinese Journal of Scientific Instrument, 2008, 29(11): 2357-2361. (in Chinese with English abstract)
[79] 楊鵬程,杜軍蘭,李燕,等. 紫外吸收光譜法結(jié)合PLS對多組分溶液中硝酸鹽濃度的測定[J]. 海洋技術(shù)學(xué)報,2013,32(2):115-119. Yang Pengcheng, Du Junlan, Li Yan, et al. Determination of nitrate in the presence of multi component solution based on ultraviolet spectra combined with PLS methods[J]. Ocean Technology, 2013, 32(2): 115-119. (in Chinese with English abstract)
[80] Song K, Li L, Tedesco L P, et al. Remote estimation of chlorophyll-a in turbid inland waters: Three-band model versus GA-PLS model[J]. Remote sensing of environment, 2013, 136: 342-357.
[81] Wang X, Fu L, Ma L. Semi-supervised support vector regression model for remote sensing water quality retrieving [J]. Chinese Geographical Science, 2011, 21(1): 57-64.
[82] Chen B, Wu H, Li S F Y. Development of variable pathlength UV–Vis spectroscopy combined with partial- least-squares regression for wastewater chemical oxygen demand (COD) monitoring[J]. Talanta, 2014, 120: 325-330.
[83] Guo Y, Liu X, Han Y, et al. Effective enrichment and simultaneous quantitative analysis of trace heavy metal ions mixture in aqueous samples by the combination of radial electric focusing solid phase extraction, UV-Vis spectrophotometric determination and partial least squares regression[J]. Water Air & Soil Pollution, 2017, 228(8): 317.
[84] Dahlén J, Karlsson S, B?ckstr?m M, et al. Determination of nitrate and other water quality parameters in groundwater from UV/Vis spectra employing partial least squares regression[J]. Chemosphere, 2000, 40(1): 71-77.
[85] Araújo M C U, Saldanha T C B, Galvao R K H, et al. The successive projections algorithm for variable selection in spectroscopic multicomponent analysis[J]. Chemometrics and Intelligent Laboratory Systems, 2001, 57(2): 65-73.
[86] Sarragu?a M C, Paulo A, Alves M M, et al. Quantitative monitoring of an activated sludge reactor using on-line UV-Visible and near-infrared spectroscopy[J]. Analytical and Bioanalytical Chemistry, 2009, 395(4): 1159.
[87] Feng X Y, Wang Q Q, Zhang J, et al. Studying aromatic compounds in infrared spectra based on support vector machine[J]. Vibrational Spectroscopy, 2007, 44(2): 243-247.
[88] Thissen U, Ustün B, Melssen W J, et al. Multivariate calibration with least-squares support vector machines[J]. Analytical Chemistry, 2004, 76(11): 3099.
[89] Choi M, Han S. Remote sensing imageries for land cover and water quality dynamics on the west coast of Korea.[J]. Environmental Monitoring & Assessment, 2013, 185(11): 9111.
[90] Sun D, Qiu Z, Li Y, et al. Detection of total phosphorus concentrations of turbid inland waters using a remote sensing method[J]. Water Air & Soil Pollution, 2014, 225(5): 1953.
[91] Tan Guohua,Yan Jianzhuo, Gao Chen. Prediction of water quality time series data based on least squares support vector machine[J]. Procedia Engineering, 2012, 31: 1194-1199.
[92] 曹泓. 基于多源光譜數(shù)據(jù)融合的水產(chǎn)養(yǎng)殖水質(zhì)有機物濃度快速檢測研究[D]. 杭州:浙江大學(xué),2014.
[93] 吳國慶,畢衛(wèi)紅. 多源光譜特征組合的COD光學(xué)檢測方法研究[J]. 光譜學(xué)與光譜分析,2014,34(11):3071-3074. Wu Guoqing, Bi Weihong. Research on chemical oxygen demand optical detection method based on the combination of multi-source spectral characteristics[J]. Spectroscopy and Spectral Analysis, 2014, 34(11): 3071-3074. (in Chinese with English abstract)
[94] Abyaneh H Z. Evaluation of multivariate linear regression and artificial neural networks in prediction of water quality parameters[J]. Journal of Environmental Health Science and Engineering, 2014, 12(1): 40.
[95] Zakaluk R, Ranjan R S. Artificial neural network modelling of leaf water potential for potatoes using RGB digital images: A greenhouse study[J]. Potato Research, 2006, 49(4): 255-272.
[96] Liu Q F, Kim S H, Lee S. Prediction of microfiltration membrane fouling using artificial neural network models[J]. Separation and Purification Technology, 2009, 70(1): 96-102.
[97] 趙煜. 基于電子舌和幾種神經(jīng)網(wǎng)絡(luò)模型的金魚養(yǎng)殖水檢測研究[D]. 杭州:浙江大學(xué),2013.
[98] Xie L J, Ye X Q, Liu D H, et al. Application of principal component-radial basis function neural networks (PC- RBFNN) for the detection of water-adulterated bayberry juice by near-infrared spectroscopy[J]. Journal of Zhejiang University Science B, 2008, 9(12): 982-989.
[99] Mesquita D P, Quintelas C, Amaral A L, et al. Monitoring biological wastewater treatment processes: Recent advances in spectroscopy applications[J]. Reviews in Environmental Science & Bio/technology, 2017, 16(3): 395-424.
[100] Mi Y P, Wang X P, Jin X. Water COD prediction based on machine learning[J]. Journal of Zhejiang University, 2008, 42(5): 790-793.
[101] Ji H, Xu Y, Li S, et al. Simultaneous determination of iron and manganese in water using artificial neural network catalytic spectrophotometric method[J]. Journal of Ocean University of China (English Edition), 2012, 11(3): 323-330.
[102] 陳明,朱文婷,周汝雁,等. BP神經(jīng)網(wǎng)絡(luò)信息融合技術(shù)在水質(zhì)監(jiān)控中的應(yīng)用[J]. 微計算機信息,2010,26(10):15-17. Chen Ming, Zhu Wenting, Zhou Ruyan, et al. The application of information fusion technology based on BP neural network in water monitoring[J]. Microcomputer Information, 2010, 26(10): 15-17. (in Chinese with English abstract)
[103] 宋協(xié)法,馬真,萬榮,等. 人工神經(jīng)網(wǎng)絡(luò)在凡納濱對蝦養(yǎng)殖水質(zhì)預(yù)測中的應(yīng)用研究[J]. 中國海洋大學(xué)學(xué)報:自然科學(xué)版,2014,44(6):28-33. Song Xiefa, Ma Zhen, Wan Rong, et al. Applicability of artificial neural network in the quality prediction of litopenaeus vannamei culturing water[J]. Periodical of Ocean University of China, 2014, 44(6): 28-33. (in Chinese with English abstract)
[104] Qu J H, Sun D W, Pu H. Vis/NIR Chemical imaging technique for predicting sodium humate contents in aquaculture environment[J]. Water Air and Soil Pollution, 2017, 228(5): 177.
Review and trend of water quality detection in aquaculture by spectroscopy technique
Li Xinxing1,2, Zhu Chenguang1, Zhou Jing1, Sun Longqing1, Cao Xiamin3, Zhang Xiaoshuan2,4※
(1.100083,; 2.100083,; 3.215200,; 4.100083,)
The water quality of aquaculture is a key factor concerning the economic benefits of aquaculture and the quality of aquatic products. In recent years, with the development of economy, the discharge of industrial wastewater and domestic sewage has greatly increased, resulting in environmental pollution, for example, the water quality of aquaculture ponds has been polluted. In order to achieve the goal of high yield and safe breeding at the same time of environmental protection and energy conservation, scholars have paid attention to the rapid and accurate acquisition of aquaculture water quality information, which was the important research content of the smart agriculture and agricultural Internet of Things. Water quality monitoring technology based on spectral analysis is an important development direction of aquaculture water quality monitoring. Compared with traditional chemical analysis, electrochemical analysis and chromatographic analysis methods, spectral analysis technology is more simple and convenient, consumes a small quantity of reagents, and is reproducible. This article summarizes and sorts the existing domestic and foreign research literatures, and systematically analyzes and discusses the important parameters of water quality monitoring, data preprocessing methods, feature band extraction, and detection model algorithms based on spectroscopy. This article reviews the COD (chemical oxygen demand) water quality monitoring methods, total nitrogen water quality monitoring methods, total phosphorus water quality monitoring methods, heavy metal water quality monitoring methods, covering traditional chemical methods and spectral analysis methods of these parameters. This article compares and analyzes the spectral method and the traditional methods. We find that compared with the traditional water quality monitoring methods, the spectral technology is non-invasive, rapid rapid monitoring, repeatable and accurate. The sensitive spectral bands of the above parameters are summarized. The data preprocessing algorithm includes Savitzky-Golay smoothing, wavelet analysis, and multivariate scatter correction, the feature band extraction algorithm includes continuous projection algorithm, no-information variable elimination algorithm, and principal component analysis, and the model includes partial least squares algorithm, least squares algorithm, and artificial neural network. The advantages, disadvantages and scopes of application of these algorithms are summarized and compared. The spectrum detection process of these algorithms is analyzed. Among them, a detailed review of the application of model algorithms in water quality monitoring is conducted, and the prediction results of each water quality prediction model algorithm are statistically analyzed. The results show that online aquaculture water quality testing will be the focus of research. Multi-parameter monitoring is the development direction of aquaculture water quality monitoring. For the processing of spectral data, the combination of multiple data processing algorithms will still dominate. Nonlinear modeling will become the mainstream method for water quality data analysis of aquaculture and will become the mainstream method for the application of spectral technology to water quality detection of aquaculture.
spectroscopy; aquaculture; water quality; monitoring model
10.11975/j.issn.1002-6819.2018.19.024
S959
A
1002-6819(2018)-19-0184-11
2018-05-04
2018-09-03
國家重點研發(fā)計劃項目(2017YFE0111200);農(nóng)村領(lǐng)域國家科技計劃資助項目(2015BAD7B-5)
李鑫星,副教授,主要研究方向為農(nóng)業(yè)系統(tǒng)與知識工程。 Email: lxxcau@cau.edu.cn
張小栓,教授,主要研究方向為農(nóng)業(yè)經(jīng)濟和信息系統(tǒng)工程。Email: zhxshuan@cau.edu.cn
李鑫星,朱晨光,周 婧,孫龍清,曹霞敏,張小栓. 光譜技術(shù)在水產(chǎn)養(yǎng)殖水質(zhì)監(jiān)測中的應(yīng)用進展及趨勢[J]. 農(nóng)業(yè)工程學(xué)報,2018,34(19):184-194. doi:10.11975/j.issn.1002-6819.2018.19.024 http://www.tcsae.org
Li Xinxing, Zhu Chenguang, Zhou Jing, Sun Longqing, Cao Xiamin, Zhang Xiaoshuan. Review and trend of water quality detection in aquaculture by spectroscopy technique[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(19): 184-194. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2018.19.024 http://www.tcsae.org