基于近红外光谱技术与BP-ANN算法的豆粕品质快速检测

应用近红外漫反射光谱技术结合误差反向传递人工神经网络(BP-ANN)算法,建立豆粕品质(包括水分、粗蛋白、残油)的定量分析模型。将豆粕漫反射吸收光谱数据进行SNV、DT、SG求导、SG平滑和均值中心化处理,然后采用偏最小二乘方法(PLS)降维获取主成分,并优化选择合适的隐含层节点数、隐含层和输出层转化函数,建立校正模型,并用验证样品对校正模型进行验证。结果显示,BP-ANN法建立的水分、粗蛋白和残油的预测相关系数(R)分别为0.981、0.988、0.982,预测标准偏差(SEP)分别为0.120、0.216、0.036,均优于PLS建模方法结果,且满足传统分析方法的重复性要求,表明BP-ANN方法可用于生产过程豆粕品质的快速监控。     

Rapid determination of tetracycline in milk by FT-MIR and FT-NIR spectroscopy

The feasibility of measuring tetracycline at the ppb levels in milk was investigated by Fourier transform mid-infrared (FT-MIR) and Fourier transform near-infrared (FT-NIR) spectroscopic techniques. Milk samples spiked with different concentrations of tetracycline were scanned using FT-MIR and FT-NIR spectroscopy. Suitable spectral wave number regions were selected for principal least square (PLS) regression models development. Prediction errors were high when the calibration model was developed using the wide range of tetracycline concentrations (4 to 2000 ppb) in milk. Maximum correlation coefficient (R2) value of about 0.89 was obtained for the validation models developed using different concentration ranges. Prediction errors were high for FT-NIR method. Results indicated that FT-MIR spectroscopy could be used for rapid detection of tetracycline hydrochloride residues in milk.     

FTIR spectroscopic characterization of soy proteins obtained through AOT reverse micelles

Fourier transform infrared (FTIR) method was used to study the conformations of soy protein obtained through aqueous solution and bis(2-ethylhexyl) sodium sulfosuccinate (AOT) reverse micelle extraction. The results showed that there were changes in signal intensity and/or position of IR bands at 4000–400 cm⁻¹ when the soy proteins were separated by two extraction methods. The FTIR spectral changes were subsequently assessed using the second derivative spectroscopy in the amide I region (1700–1600 cm⁻¹). The contents of α-helix, β-sheet, turn and irregular conformations for soy proteins using aqueous solution extraction were 11.6%, 32.8%, 44.3% and 11.3%, respectively; while using AOT reverse micelle extraction 13.1%, 41.6%, 32.5% and 12.9%, respectively. The amount change of these structures might affect functional properties of soy proteins.     

Performance of Error Back Propagation vis-à-vis Radial Basis Function Neural Network: Part II: Reverse Engineering of Woven Fabrics

Abstract

This paper describes the application of an artificial neural network approach to engineer the design of woven wool and wool blended suiting fabric, to be used by the weavers. Two neural network models based on error back propagation and radial basis function algorithms are used for ascertaining fabric constructional parameters such as fibre composition, yarn density, yarn tex, weave, yarn crimp and yarn twist to obtain desired low stress mechanical properties and other properties such as breaking strength and extension, bending rigidity, shear rigidity and tear strength of the suiting fabric. Of the two networks, radial basis function network is again found to be fast to train and easier to design than back propagation network. Evaluation of both the models for each fabric property specification shows good agreement between predicted and generally accepted fabric, yarn structure–property relationships.     

基于近红外化学成分特征吸收光谱的野生台蘑鉴别模型研究

本文建出一种应用近红外光谱技术鉴别野生台蘑的新方法。使用FieldSpec3便携式近红外光谱仪对包括野生台蘑在内的13种蘑菇进行漫反射光谱采集。将采集的数据经小波去噪后,对可见与近红外光谱(350～2500nm)进行峰谷筛选,所得峰谷集经主成分分析降维,取方差贡献率大于99.9%的5个主成分作为BP神经网络的输入值,建立数学模型。该模型在偏差±0.05内,对未知样本正确识别率为100%。本结果表明利用近红外漫反射光谱可以很好地鉴别野生台蘑。     

基于傅里叶变换红外光谱的食用油质量安全检测技术研究进展

为了进一步拓展傅里叶变换红外（Fourier transform infrared,FTIR）光谱技术在食用油分析和检测中的应 用范围,本文综述了近年来FTIR光谱技术在食用油品质分析和安全检测方面的研究进展。其中品质分析包括理化 指标（过氧化值、反式脂肪酸含量、游离脂肪酸含量、水分体积分数、碘值、羰基值及多指标分析和检测）和氧化 稳定性;安全检测即食用油真伪鉴别。通过对光谱采集、光谱范围选择、数据处理及基底效应消除等方面进行分析 和探讨,以期为FTIR光谱技术在食用油质量安全检测方面的进一步应用提供参考。     

光谱预处理对太赫兹光谱预测猪肉K值的影响

采用太赫兹（terahertz,THz）光谱分析技术无损检测猪肉的新鲜度K值,但水会强烈吸收THz波,从而严重影响THz波对肉的检测。考察预处理方法对削弱水的干扰、提升THz光谱检测猪肉K值的模型性能的影响。分别采用多元散射校正、标准正态变量变换、一阶微分、二阶微分4?种预处理方法对衰减全反射光谱进行预处理,基于反向传播人工神经网络回归算法建立猪肉K值的THz光谱预测模型,比较研究4?种预处理方法后的模型预测精度。研究表明：一阶微分预处理方法效果最好,能够消除光谱基线漂移,提高光谱质量。与原始光谱相比,模型的预测集相关系数（Rp）从0.34提高到0.75,预测集均方根误差从20.24%降低到14.36%。因此,选择合适的光谱预处理技术对提高模型预测精度非常重要,采用一阶微分预处理后的THz光谱数据建立反向传播人工神经网络模型能够无损检测猪肉的新鲜度K值。     

FTIR spectroscopic characterization of irradiated hazelnut (Corylus avellana L.)

Radiation induced molecular changes in macromolecular components of hazelnut tissues were investigated by mid-Fourier transform infrared (FTIR) spectroscopy. Irradiation dose of 1.5 kGy (low) and 10 kGy (high) were applied. The changes in frequency, signal intensity and intensity ratio of IR bands revealed that the unsaturated lipid concentration increased for low dose treated samples whereas it decreased and peroxidation appeared at high dose treatment. The low dose irradiation treatment, slightly increased the total lipid content whereas it dramatically decreased for high dose treatment. A slight increase in the lipid to protein ratio was observed for low dose treatment, whilst this ratio significantly decreased for high dose treatment. In addition, the high dose γ-irradiation caused alterations in the structure of hazelnut proteins, as cross-linking and aggregation occured in protein molecules. These results indicate that FTIR spectroscopy can be successfully used to monitor food irradiation.     

Molecular identification of marine yeast and its spectroscopic analysis establishes unsaturated fatty acid accumulation

Marine microbes are competent organisms, some of which can accumulate large amounts of lipids. A yeast strain, Rhodotorula mucilaginosa AMCQ8A was isolated from the marine water of the Queenscliff region, Victoria, Australia. The yeast isolate was identified by sequencing 18s rDNA genes. Scanning electron microscopy images revealed scars on the surface of the yeast cells. The Fourier transform infrared spectroscopy microspectroscopy studies demonstrated the presence of unsaturated fatty acids by differential microscopic analysis. The sharp band at 1745?cm(-1) was represented by ν(CO) stretches of ester functional groups from lipids and fats, and therefore indicated the presence of total lipids produced by the cells. Over 65% of the fatty acids from the yeast strain were analyzed as C(16) and C(18:1) with omega-3 content from about 6% to 7%. Thus, this marine-derived yeast could be a potential source of lipids, including omega-3 fatty acids.     

有机溶剂和混合油脂对脂肪酶活性恢复的影响

本文通过多次催化酯交换方法,研究了最优催化条件下脂肪酶活性和结构的关系,以及低活性脂肪酶的活性再生技术.在催化合成1,3-二油酸-2-棕榈酸甘油三酯的研究中,利用荧光光谱和傅里叶变换红外光谱,分析最优条件下不同使用次数脂肪酶的三级与二级结构,研究脂肪酶活性变化与其二级、三级结构之间的关系.使用异丙醇和混合油脂处理脂肪酶...     

近红外光谱-人工神经网络的模型优化用于银耳产地识别研究

研究通过近红外光谱技术（NIRS）结合人工神经网络技术（ANN）识别银耳的不同产地。实验以四川省与福建省两个产地共120组银耳样品为研究对象,对其进行近红外光谱测定,计算光谱吸收值的平均偏差与一阶导数进而选取有效数据,结合主成分分析方法将原始数据降维并采用反向人工神经网络技术构建近红外分析模型。结果显示,通过对有效数据主成分分析,前3个主成分的累计方差贡献率达到100%,判断准确率为88.3%;进一步采用人工神经网络优化模型,在输出层为2隐藏层为11时,判断准确率达100%;此时校正集与预测集的均方根误差分别为3.05×10^-2与2.90×10^-2,模型具有良好的泛化能力。因此,结合人工神经网络的近红外光谱检测技术,优化检测模型,能够准确、快速地识别银耳产地,为食品原材料的质量控制及地理标志的建立提供科学依据。      

基于涂膜法FTIR的食用油羰基值快速检测

为了建立食用油羰基值快速检测方法,以菜籽油为实验材料,茴香醛为羰基标准物配制系列羰基值（0～60 mmol/kg）油样。用涂膜法采集油样光谱,建立羰基值与特征吸收峰1 703/1 570 cm－1处（1 703 cm－1处峰高两点基线均为1 570 cm－1,下同）吸光度的关系。然而,不同种类零羰基值的食用油在特征吸收峰1 703 cm－1处的吸收不相同而引起基底效应。因此,利用二维相关振动光谱技术寻找特征吸收峰的相关吸收峰,以校正食用油间的基底效应。结果表明：油样在1 703 cm－1处的吸光度与羰基值高度线性相关,利用相关吸收峰3 472 cm－1校正后的模型为y=0.005 4x＋0.090 9;相关系数（R）为0.991 9,标准偏差（s）为0.013 mmol/kg。模型预测值与国标法测定值高度线性相关,所得方程的斜率接近于1,R值均在0.99以上,两种方法测定结果十分接近。表明涂膜法傅里叶变换红外光谱检测食用油羰基值是可行的。     

近红外光谱法鉴别奶牛饲料中三聚氰胺甲醛树脂的可行性研究

收集国内常用的、具有代表性的奶牛精补料33个样品,制备99个三聚氰胺甲醛树脂(MF)掺假样品,在全光谱范围内进行近红外透反射光谱扫描,选择合适的前处理方法,采用BP神经网络方法和PLS-LDA方法分别建立判别模型。建立的BP神经网络判别分析模型的预测正确率为100%,建立的PLS-LDA判别分析模型的交互验证最低错误率为0.0778,模型错分率为0.0667,模型预测错误率为0.1429。说明利用近红外透反射光谱建立定性分析模型来检测奶牛饲料中是否掺有MF的研究是可行的。     

人工神经网络及其在盐田水体遥感图像分类中的研究

卫星遥感技术的发展为盐业生产资源监测提供了有效的手段.文章首先介绍了遥感图像分类的原理及其常用方法,接着就人工神经网络基本概念和常用算法进行阐述,并着重介绍了反向传播网络(BP网络),最后以天津塘沽盐场区域为研究对象,用BP网络进行了盐田水体识别实验.     

基于HVI指标与RBF神经网络的纱线质量预测

将人工神经网络与HVI指标相结合,提出了一种基于RBF神经网络的纱线质量预测方法,克服了BP神经网络训练效率低、容易陷入局部极小化等不足。通过实例分析,利用马克隆值等14项HVI指标和工艺参数对纱线的单纱断裂强度等4项指标进行预测,试验结果表明:与传统纱线质量预测方法相比,基于RBF神经网络的纱线质量预测模型表现出了优良的预测精度及稳定性,对降低生产成本、提高生产效率具有积极意义。     

基于近红外光谱和共轭梯度神经网络的板栗褐变检测

为了实现板栗褐变的无损检测,本实验以     

Effect of high pressure treatment on aggregation and structural properties of soy protein isolate

The aggregation and structural properties of soy protein isolate (SPI), induced by high pressure (HP) treatment at 200-600 MPa were investigated by size-exclusion chromatography combined with multi-angle laser scattering (SEC-MALLS) and fourier transform infrared (FTIR) spectroscopy. HP treatment at lower pressure level (e.g., 200 MPa) resulted in formation of marked insoluble aggregate of SPI, while the treatment at higher pressure level (e.g., 600 MPa) led to transformation of insoluble aggregate to soluble one. The soluble aggregate formed at 400 or 600 MPa had much less mean molecular weight (MW) (about 5.2 × 10⁶ g/mol) than that at 200 MPa (about 1.6 × 10⁷ g/mol), and was also much more homogenous in MW distribution. FTIR analyses confirmed changes in secondary and tertiary structures, induced by HP treatment. These results can provide direct evidence or explanation for HP-induced modification of soy proteins.     

近红外漫反射光谱法快速检测胶原肽粉中低聚肽含量

目的建立近红外漫反射光谱法(near infrared reflectance spectroscopy, NIRS)快速检测胶原肽粉核心功效成分低聚肽含量的分析方法。方法基于径向基神经网络(radial basis function neural network, RBFNN)的方法,分析比较了多元散射校正(multiplescattercalibration,MSC)、标准正态变换(standardnormalvariation,SNV)的光谱预处理方法,分别结合了主成分分析(principalcomponentanalysis,PCA)、独立成分分析(independent component analysis, ICA)的方法提取特征信息,优化输入变量、扩展系数等参数,建立了低聚肽含量的检测模型,同时为了比较方法优劣,也建立了相应的偏最小二乘方法(partial least squares, PLS)模型。结果非线性建模方法 RBFNN比线性PLS方法模型效果更好,与常用的PCA-RBFNN模型相比,本研究采用的ICA-RBFNN模型结果更准确,模型独立验证集的相关系数R是0....