摘 要：高光譜成像與近紅外光譜（near infrared spectroscopy，NIR）技術(shù)是現(xiàn)代食品檢測領(lǐng)域的重要手段，本研究對2 種技術(shù)在雞肉品質(zhì)無損檢測中的預(yù)測精度進行研究。選用62 份新鮮程度不同的雞胸肉，提取其高光譜感興趣區(qū)域（region of interest，ROI）的光譜曲線，并測定樣品的揮發(fā)性鹽基氮（total volatile base nitrogen，TVB-N）含量和菌落總數(shù)（total viable count，TVC），利用OPUS 6.0光譜處理軟件搜尋最佳的光譜預(yù)處理和波段組合，分別建立2 個指標的偏最小二乘法（partial least square，PLS）定量分析模型。NIR樣本選用30 份新鮮程度不同的雞胸肉，測定其TVB-N含量和TVC，建立PLS的交叉驗證模型。結(jié)果表明：利用高光譜的ROI平均光譜建立的TVB-N含量與TVC模型的相關(guān)系數(shù)（R2）分別為0.965和0.919，均方根誤差（root mean square error of cross validation，RMSECV）分別為0.121和0.215；利用NIR建立的TVB-N含量與TVC預(yù)測模型的R2分別為0.801和0.780，RMSECV分別為0.232和0.312。由此可見，基于高光譜的ROI區(qū)域光譜建立的預(yù)測模型在雞肉品質(zhì)無損檢測中具有比NIR更高的預(yù)測精度。

關(guān)鍵詞：雞肉新鮮度；高光譜成像；偏最小二乘法；近紅外光譜

Abstract： Hyperspectral imaging and near infrared spectroscopy （NIR） are two important techniques in modern food detection. This study intended to study the prediction accuracy of the two techniques for non-destructive chicken quality detection. Totally 62 chicken breast samples with different freshness were selected for hyperspectral imaging. Spectral data were extracted from the region of interest （ROI）. Total volatile base nitrogen （TVB-N） content and total viable count （TVC） were measured. The optimal combination of spectral pretreatment and band was searched by OPUS software （version 6.0）. A predictive model to quantify TVB-N and TVC was established by means of partial least squares （PLS） regression， respectively. Moreover， another 30 samples with different freshness were used to develop a PLS model for predicting TVB-N content and TVC by NIR spectroscopy， respectively. The performance of each model was evaluated using

cross-validation. The results showed that the correlation coefficients （R2） of the TVB-N content and TVC prediction models developed from the ROI average spectra from hyperspectral images were 0.965 and 0.919 with a root mean square error of cross validation （RMSECV） of 0.121 and 0.215， respectively， while those of the prediction models established from NIR spectra were 0.801 and 0.780 with a RMSECV of 0.232 and 0.312， respectively. It can be concluded that the model based on ROI spectra from hyperspectral images has higher prediction accuracy for chicken quality compared with the NIR model.

Key words： chicken freshness； hyperspectral imaging； partial least squares （PLS）； near infrared spectroscopy （NIR）

DOI：10.7506/rlyj1001-8123-201712006

中圖分類號：TS251.1 文獻標志碼：A 文章編號：1001-8123（2017）12-0030-06

引文格式：

邢素霞， 王睿， 郭培源， 等. 高光譜成像及近紅外技術(shù)在雞肉品質(zhì)無損檢測中的應(yīng)用[J]. 肉類研究， 2017， 31（12）： 30-35. DOI：10.7506/rlyj1001-8123-201712006. http：//www.rlyj.pubendprint