摘要
以豆粕和3种抗生素菌渣为研究对象,通过傅里叶变换近红外显微成像系统采集样品近红外显微图像;对采集到的近红外显微图像进行光谱重构,并对所有样品光谱进行预处理,利用Duplex算法分别从不同的样品预处理光谱中筛选具有代表性的光谱建立豆粕和抗生素菌渣的特征光谱库。使用偏最小二乘判别分析(PLS-DA)与支持向量机判别分析(SVM-DA)结合不同的光谱预处理方法,构建豆粕与不同种类抗生素菌渣的近红外显微成像定性判别模型。结果表明:构建的2种模型均能有效对试验中所用豆粕和抗生素菌渣样品进行鉴别分析,正确率均在99.4%以上。进一步比较研究发现,一阶导数+SNV的预处理方式优于无预处理、一阶导数、二阶导数;SVM-DA的模型效果优于PLS-DA,SVM-DA中特征提取方法 PLS优于PCA。
AMR（ antibiotic mycelial residue） added to animal feed easily leads to drug resistance influencing human health and environment. However,there is a lack of effective detection methods,especially fast and convenient detection technology,to distinguish AMR from animal feed. In order to search effective detection methods,qualitative discriminant analysis of soybean meal and antibiotic residue was made at first. The feasibility of near infrared micro-imaging for the identification of soybean meal and antibiotic mycelial residues was explored. Three soybean meal samples and three kinds of antibiotic mycelial residues were used to collect the near-infrared microscopic images of the samples by Fourier transform near-infrared microscopy. The near-infrared microscopic images collected were reconstructed and the spectra of all the samples were pretreated. The Duplex algorithm was employed to screen the representative spectra from pretreatment spectra of different samples to establish spectral library of soybean meal and antibiotic mycelial residues. Different discriminant models of soybean meal and different kinds of antibiotic mycelial residues were built by using different pretreatment methods combined with PLS-DA（ partial least squares discriminant analysis） and SVM-DA（ support vector machine discriminant analysis）. The results showed that two kinds of modeling methods based on near-infrared micro-imaging spectroscopy were effective in the identification of three kinds of antibiotic mycelial residues and soybean meal samples, and the correctness rate was above 99. 4%. The first-order derivative ＋ SNV preprocessing method was better than that without preprocessing,the first derivative and the second derivative. SVM-DA model was superior to PLS-DA,and SVM-DA in feature extraction method was better than PCA（ principal component analysis）. The results presented indicated that the near infrared microscopic imaging technique can be used to qualitatively distinguish antibiotic mycelialresidue from soybean meal,and it also provided theoretical basis for further research.
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2017年第12期363-369,共7页
Transactions of the Chinese Society for Agricultural Machinery
基金
中国农业科学院基本科研业务费专项(1610072017001)和中国农业科学院"饲料质量安全检测与评价"创新团队经费项目
关键词
豆粕
抗生素菌渣
鉴别分析
近红外显微成像
soybean meal
antibiotic mycelial residue
discriminant analysis
near infrared microimaging
