【摘要】：以豆粕和3種抗生素菌渣為研究對象,通過傅里葉變換近紅外顯微成像系統(tǒng)采集樣品近紅外顯微圖像;對采集到的近紅外顯微圖像進行光譜重構(gòu),并對所有樣品光譜進行預處理,利用Duplex算法分別從不同的樣品預處理光譜中篩選具有代表性的光譜建立豆粕和抗生素菌渣的特征光譜庫。使用偏最小二乘判別分析(PLS-DA)與支持向量機判別分析(SVM-DA)結(jié)合不同的光譜預處理方法,構(gòu)建豆粕與不同種類抗生素菌渣的近紅外顯微成像定性判別模型。結(jié)果表明:構(gòu)建的2種模型均能有效對試驗中所用豆粕和抗生素菌渣樣品進行鑒別分析,正確率均在99.4%以上。進一步比較研究發(fā)現(xiàn),一階導數(shù)+SNV的預處理方式優(yōu)于無預處理、一階導數(shù)、二階導數(shù);SVM-DA的模型效果優(yōu)于PLS-DA,SVM-DA中特征提取方法 PLS優(yōu)于PCA。
[Abstract]:Taking soybean meal and three kinds of antibiotic bacteria dregs as research objects, the near infrared microscopic images of samples were collected by Fourier transform near infrared microscopic imaging system, and the acquired near infrared microscopic images were reconstructed by spectral analysis. All the sample spectra were pretreated and the characteristic spectrum library of soybean meal and antibiotic residue was established by Duplex algorithm from different sample pretreatment spectra. Using partial least squares discriminant analysis (PLS-DA) and support vector machine discriminant analysis (SVM-DA) combined with different spectral pretreatment methods, a qualitative model of near-infrared microscopic imaging of soybean meal and antibiotic residue was established. The results showed that the two models could be used to identify and analyze soybean meal and antibiotic bacteria residue effectively, and the accuracy was over 99.4%. Furthermore, it is found that the pretreatment of first derivative SNV is better than that of no pretreatment, and the model effect of first derivative and second derivative SVM-DA is better than that of PLS-DA-SVM-DA.
【作者單位】： 中國農(nóng)業(yè)大學工學院;中國農(nóng)業(yè)科學院農(nóng)業(yè)質(zhì)量標準與檢測技術(shù)研究所;
【基金】：中國農(nóng)業(yè)科學院基本科研業(yè)務(wù)費專項(1610072017001)和中國農(nóng)業(yè)科學院“飼料質(zhì)量安全檢測與評價”創(chuàng)新團隊經(jīng)費項目
【分類號】：O657.3;S816

【相似文獻】

相關(guān)期刊論文 前2條

1 趙楊l，

本文編號：2136540