【作者】 吕程序；

【导师】 韩鲁佳；

【作者基本信息】 中国农业大学 ， 农业工程， 2014， 博士

【摘要】 为监控饲料品质,保障饲料质量安全,本文通过二维相关光谱技术、多尺度近红外光谱分析及其融合联用方法,对蛋白饲料原料进行快速、无损判别的方法研究。本研究对丰富我国蛋白饲料原料检测手段具有重要的理论意义和实用价值。研究主要内容及结果如下：以配合饲料和肉骨粉为研究对象,进行基于二维相关变量优选的显微近红外光谱高效判别的研究。研究提出了一种基于二维相关近红外光谱自动峰/交叉峰选择敏感变量的方法,并基于此发现配合饲料/肉骨粉存在12个敏感近红外变量：6852、6388、6320、5788、5600、5244、4900、4768、4572、4336、4256和4192cm-1。基于此构建了配合饲料/肉骨粉显微近红外光谱的判别分析方法,结果显示该方法可指派敏感变量归属,判别分析效果良好(判断正确率大于99%),数据压缩效率高(99%),分析效率较全谱提高(大于14%),选择的变量传递性好。以鱼粉和肉骨粉为研究对象,进行二维相关变量优选的显微近红外光谱高效判别的研究。研究提出了一种基于二维相关显微近红外光谱自动峰选择敏感变量的方法,并基于此发现鱼粉/肉骨粉存在11个显微近红外敏感变量：5788、5748、5676、4900、4468、4368、4336、4300、4264、4232和4196cm-1。基于此构建了鱼粉/肉骨粉显微近红外光谱的判别分析方法,结果显示该方法可指派敏感变量归属,判别分析效果良好(判断正确率大于98%),数据压缩效率高(99%),分析效率较全谱提高(大于67%)。以鱼粉和肉骨粉为研究对象,进行温扰二维相关近红外光谱可视化判别研究。研究发现在一定的温度扰动(20-60℃)下,鱼粉/肉骨粉二维相关近红外同步谱存在各自特异性指纹特征,基于此构建了相关的鱼粉/肉骨粉二维相关近红外光谱可视化判别方法,即：6000-5400cm-1范围的动态谱经基线校正+小波变换预处理和二维相关同步分析。该方法机理清晰,直观可视,有助于提高模型共享性。以不同种属肉骨粉为研究对象,进行温扰二维相关近红外光谱可视化判别研究。研究发现在一定的温度扰动(10-65℃)下,鸡/牛/猪源肉骨粉的二维相关近红外同步谱存在各自特异性指纹特征,基于此构建了相关的不同种属肉骨粉二维相关近红外光谱可视化判别方法,即：8600-8200cm-1范围的动态谱经基线校正预处理、基于主成分分析的光谱重构和二维相关同步分析；5600-5500cm-1范围的动态谱经二阶导数预处理、基于主成分分析的光谱重构和二维相关同步分析。该方法直观可视,有助于提高模型共享性。以鱼粉和豆粕为研究对象,进行基于样本-样本二维相关鱼粉/豆粕近红外光谱的快速判别研究。基于一阶导数预处理的样本-样本二维相关同步切线谱可提取区分鱼粉/豆粕的数值特征,基于此确定的判别阈值可用于鱼粉/豆粕的快速定性判别,验证集样本判断正确率为100%。研究提出了一种基于样本-样本二维相关近红外光谱的鱼粉/豆粕快速判别方法,该方法算法简单,识别速度快。更多还原

【Abstract】 To monitor the feed quality and ensure the feed safety, this research focuses on the two-dimensional correlation spectroscopy, variable-scale near-infrared spectroscopy and their combination technique to effectively discriminate protein feed material. The research work is of important scientific and practical value in enriching rapid detection method of protein feed. The main research contents and conclusions of are as follows:Effective near-infrared microscopic discrimination of compound feed and meat and bone meal has been investigated via a variable selection by two-dimensional correlation spectroscopy. The variable selection method is proposed based on autopeaks and cross peaks of two-dimensional correlation synchronous.12main sensitive variables are identified, and they are6852,6388,6320,5788,5600,5244,4900,4768,4572,4336,4256and4192cm-1. These sensitive variables are then used to build a discrimination model, which yields a correct classification of99%. This method presents the advantages of introducing an interpretive aspect to variable selection; enabling efficient compression and analysis of spectral data; and enabling result transferability to discriminant analysis.Effective near-infrared microscopic discrimination of fishmeal and meat and bone meal has been investigated via a variable selection by two-dimensional correlation spectroscopy. The variable selection method is proposed based on autopeaks of two-dimensional correlation near-infrared microscopy.11main sensitive variables are identified, and they are5788,5748,5676,4900,4468,4368,4336,4300,4264,4232and4196cm-1. These sensitive variables are then used to build a discrimination model, which yields a correct classification more than98%. This method presents the advantages of introducing an interpretive aspect for variable selection, enabling efficient compression and analysis of spectral data.Visualized recognition of fishmeal and meat and bone meal has been investigated via temperature-dependent two-dimensional correlation near-infrared spectroscopy. Fingerprint visual pattern differences are observed from20-60℃temperature-dependent synchronous maps between fishmeal and meat and bone meal. Its visual discrimination method is proposed:the two-dimensional correlation synchronous calculated from dynamic spectra at the range of6000-5400cm-1with the pretreatment of baseline correction and wavelet transform. This method presents the advantages of clearly clarifying the discrimination mechanism; enabling visualized and intuitive recognition; and improving sharing of the model.Visualized recognition of different specific meat and bone meal has been investigated via temperature-dependent two-dimensional correlation near-infrared spectroscopy. Fingerprint visual pattern differences are observed from10-65℃synchronous maps between poultry, bovine and porcine originated meat and bone meal. The visual discrimination methods are proposed:the two-dimensional correlation synchronous calculated from dynamic spectra at the range of8600-8200cm-1with the baseline correction pretreatment and the spectral reconstruction; synchronous calculated from dynamic spectra at the range of6000-5500cm-1with the second derivative pretreatment and the spectral reconstruction. This method presents the advantages of enabling visualized and intuitive recognition; and improving sharing of the model.Rapid identification of fishmeal and soybean meal has been investigated via a sample-sample two-dimensional correlation near-infrared spectroscopy. Numerical information for differentiating fishmeal and soybean meal is extracted by synchronous slice spectrum, it is further developed for qualitative discrimination threshold. Validation samples are100%correctly classified by these thresholds. Sample-sample two-dimensional correlation near-infrared spectroscopy is capable to discriminate fishmeal and soybean meal samples, and it presents the advantages of enabling simple computation and rapid identification.更多还原