發(fā)布時(shí)間：2018-08-21 12:40

【摘要】：為了實(shí)現(xiàn)豬肉營(yíng)養(yǎng)組分(脂肪和蛋白質(zhì))的快速、無(wú)損、實(shí)時(shí)檢測(cè),基于近紅外反射光譜設(shè)計(jì)了便攜式豬肉營(yíng)養(yǎng)組分無(wú)損檢測(cè)裝置。硬件部分包括光譜采集單元、光源單元和控制單元,并開發(fā)了相應(yīng)的檢測(cè)軟件,實(shí)現(xiàn)樣品光譜信息的有效獲取和實(shí)時(shí)分析。為了建立穩(wěn)定可靠的預(yù)測(cè)模型,考察了波段選擇、樣本分組方式和篩選變量方法對(duì)模型的影響。分別基于可見/短波近紅外(Vis/SWNIR)、長(zhǎng)波近紅外(LWNIR)及Vis/SWNIR-LWNIR,利用隨機(jī)選擇法(RS)、Kennard-Stone法(KS)和基于聯(lián)合X-Y距離的樣本劃分法(SPXY)對(duì)樣本進(jìn)行劃分,建立了脂肪和蛋白質(zhì)質(zhì)量分?jǐn)?shù)的偏最小二乘預(yù)測(cè)模型。結(jié)果發(fā)現(xiàn),基于Vis/SWNIR-LWNIR波段,利用SPXY算法進(jìn)行樣本分組,取得了最佳的預(yù)測(cè)模型。在此基礎(chǔ)上,比較分析競(jìng)爭(zhēng)性自適應(yīng)加權(quán)算法、隨機(jī)蛙跳算法和蒙特卡羅無(wú)信息變量消除-連續(xù)投影算法3種算法篩選變量建立的模型效果�；诟�(jìng)爭(zhēng)性自適應(yīng)加權(quán)算法篩選變量的模型結(jié)果最佳,對(duì)脂肪和蛋白質(zhì)建立的模型驗(yàn)證集相關(guān)系數(shù)分別為0.950 5和0.951 0。結(jié)果表明:基于近紅外反射光譜設(shè)計(jì)的便攜式豬肉組分檢測(cè)裝置可以對(duì)脂肪和蛋白質(zhì)含量進(jìn)行快速、無(wú)損、實(shí)時(shí)檢測(cè)。
[Abstract]:In order to realize the fast, nondestructive and real-time detection of pork nutrient components (fat and protein), a portable nondestructive detection device for pork nutrient components was designed based on near-infrared reflectance spectroscopy (NIR). The hardware is composed of spectral acquisition unit, light source unit and control unit, and the corresponding detection software is developed to realize the effective acquisition and real-time analysis of the sample spectral information. In order to establish a stable and reliable prediction model, the effects of band selection, sample grouping and screening variables on the model were investigated. Based on visible / short wave near infrared (Vis/SWNIR), long wave near infrared (LIR) and Vis/ SWNIR-LWNIRs, samples were divided by random selection method (RS) Kennard-Stone method (KS) and sample partition method (SPXY) based on joint X-Y distance, respectively. A partial least square prediction model for the mass fraction of fat and protein was established. The results show that the best prediction model is obtained by using SPXY algorithm to group samples based on Vis/SWNIR-LWNIR band. On this basis, the model effects of three kinds of algorithms, namely competitive adaptive weighting algorithm, stochastic leapfrog algorithm and Monte Carlo non-information variable cancellation-continuous projection algorithm, are compared and analyzed. The model result based on competitive adaptive weighting algorithm is the best. The correlation coefficient of model verification set for fat and protein is 0.950 5 and 0.951 0 respectively. The results show that the portable pork component detection device based on near infrared reflectance spectroscopy can be used to detect fat and protein content quickly, nondestructive and in real time.
【作者單位】： 中國(guó)農(nóng)業(yè)大學(xué)工學(xué)院;國(guó)家農(nóng)產(chǎn)品加工技術(shù)裝備研發(fā)分中心;
【基金】：國(guó)家重點(diǎn)研發(fā)計(jì)劃項(xiàng)目(2016YFD0401205) 公益性行業(yè)(農(nóng)業(yè))科研專項(xiàng)(201003008)
【分類號(hào)】：O657.33;TS251.51

【相似文獻(xiàn)】

相關(guān)期刊論文 前1條

1 Cynthia BOSNAK;Ewa PRUSZKOWSKI;;電感耦合等離子體質(zhì)譜檢測(cè)食物中的有毒、主要和營(yíng)養(yǎng)組分[J];生命科學(xué)儀器;2011年05期

，

本文編號(hào)：2195767