六种金属离子对轮梗霉生长及产花生四烯酸的影响

研究工业发酵中常见的Ca2 、Zn2 、Mn2 、Fe3 、Cu2 及Co2 对轮梗霉生长及产花生四烯酸的影响。结果表明:Ca2 浓度在10-4～10-2mol/L范围内促进菌体生长,但抑制AA产量。Mn2 浓度为5×10-4mol/L时AA产量达到最大,但对菌体生长都有抑制作用。Zn2 在3×10-6mol/L浓度时AA产量达最大,而生物量的变化不大。Fe3 浓度为5×10-5mol/L时,生物量达最大,而AA产量受其抑制,Cu2 浓度为5×10-6mol/L时菌体生物量和AA产量同时达到最大。Co2 在10-7mol/L浓度时生物量及AA产量同时达最大。     

基于BP神经网络的棉花颜色级预测

为了方便快捷检验棉花颜色级,采用BP神经网络算法构建了基于棉纤维色度参数的棉花颜色级预测模型。模型以棉纤维的色度参数明度值和黄度值为输入变量,以棉花颜色级为输出变量,采用三层网络,网络的拓扑结构为2-12-1。通过对280份棉花试样的颜色级进行测试和BP神经网络模型预测,结果表明:BP神经网络预测值与实测值的平均相对误差为3.203%,所建立的回归方程拟合优度达到99%。认为:所建立BP神经网络模型具有较好的预测效果。     

基于卷积神经网络的蓝印花布纹样基元分类

为更好地数字化传承与创新传统的蓝印花布纹样,并能单独提取构成纹样的图案基元并进行分类,提出一种基于卷积神经网络的纹样基元分类方法。首先,对采集的128张蓝印花布图像进行纹样基元提取,形成图像样本库,共21 212张。其次,从库中随机选取80%的图像样本作为训练集,20%作为测试集,利用5×5卷积核对训练样本进行卷积操作,将得到的特征图进行池化。通过3层卷积、3层池化及2层全连接层计算后,利用Softmax分类器得到12种分类结果。最后,通过基元样本的学习获取最佳网络模型参数,并取得较理想的分类结果。结果表明:提出的卷积神经网络模型对12种纹样基元的平均分类准确率达99.61%,检测平均准确率达98.5%,为蓝印花布纹样的研究提供了新思路。     

基于CNN神经网络的小麦不完善粒高光谱检测

利用高光谱成像技术对小麦不完善粒进行无损检测。以932个小麦为样本,其中正常粒样本486个、破损粒样本170个、虫蚀粒样本149个及黑胚粒样本127个为研究对象,通过高光谱图像采集系统采集样本的光谱信息,然后从每个样本的116个波段中选取30个波段,建立基于深度学习的卷积神经网络(convolutional neural networks,CNN)模型。实验中的CNN采用2个卷积层,第1层采用大小为3×3的32个卷积核,第2层采用大小为5×5的64个卷积核,池化层采用最大池,激活函数采用修正线性单元,为避免过拟合,在全连接层后面接入dropout层,参数设置为0.5,其他卷积参数均为默认值,得到校正集总识别率为100.00%,测试集总识别率为99.98%。最后,以支持向量机(support vector machine,SVM)为基线模型进行对比,从116个波段中选取90个波段进行建模,测试集总识别率为94.73%。通过实验对比可以看出,CNN模型比SVM模型识别率高。研究表明CNN模型能够实现对小麦不完善粒的准确、快速、无损检测。     

基于大米特征矿物元素产地鉴别建模方法比较研究

基于前期实验筛选到的与产地和母质土壤直接相关的23种特征矿物元素作为产地判别特征指标,比较Fisher判别模型和前馈神经网络预测模型的适用性。以连续3年随机采集的五常、查哈阳和建三江地理保护区274份样本作为建模对象,以模型的判别率为指标,采用电感耦合等离子体质谱仪(ICP-MS)和X射线荧光光谱仪测定样本中特征矿物元素含量(Mg、Ca、Cr、Mn、Zn、As、Rb、Sr、Ag、Cd、Sb、Te、Ba、La、Nd、Sm、Gd、Dy、Ho、Er、Yb、Pb、U),结合线性(一般线性判别分析)和非线性(前馈神经网络训练方法)模型构建方法用于产地鉴别。结果表明,Fisher判别分析模型对训练集判别率为81.5%,交叉检验判别率为78.8%,测试集总体判别率为87.5%。前馈神经网络预测模型对大米产地识别结果平均相对误差值为17.14%,产地的整体识别准确率为100%。筛选到的特征矿物元素携带了足够多的产地信息,通过前馈神经网络法建立的判别模型具有更优的判别能力,能解决小距离相似自然环境产地样本难以识别的问题。     

CV法测试电阻率的稳定性研究

本文采用化学气相沉积(CVD)方法,利用平板式外延炉,在5英寸111晶向,2×10-3Ω·cm重掺As衬底上,生长N/N+型硅外延片;试验中采用电容-电压方法,利用汞探针CV测试仪,通过金属汞与硅外延片表面接触形成肖特基势垒,测试势垒电容并转换为外延层载流子浓度和电阻率,实现对外延层电阻率的测量。在此过程中,我们采用H2O2水浴处理硅外延片表面,形成10~15?的氧化层,并对比分析了氧化温度、氧化反应时间、H2O2浓度对电阻率测试结果的影响。     

多层前馈神经网络几种算法的样本顺序敏感性

检测神经网络4种算法(有弹回的算法、OSS算法、LM算法和Regularization算法)的分类能力.结果表明,多层前馈神经网络的分类能力与算法和学习样本顺序有关;OSS算法和LM算法训练的网络分类能力对学习样本顺序最为敏感,且容易产生输出奇异点和宽幅输出波动,Regularization算法训练的网络分类能力对学习样本顺序不敏感;下凸学习样本顺序的OSS算法分类精度最高,网络分类能力最强;递增学习样本顺序的LM算法分类精度最低,网络分类能力最差;不同算法的网络分类能力对学习样本的敏感程度取决于网络权值、阈值的冗余与传递函数的性质.     

近红外光谱技术快速检测腊肉酸价和过氧化值

探讨应用傅里叶近红外光谱技术快速定量检测腊肉酸价和过氧化值的方法。腊肉样品经粉碎、混匀后在AntarisⅡ傅里叶近红外光谱分析仪上扫描,获得其近红外光谱与国标法测定的酸价和过氧化值含量数据进行关联,用傅里叶变换近红外光谱技术结合偏最小二乘法建立近红外光谱与腊肉酸价和过氧化值含量的数学模型并进行预测。结果表明：酸价模型中,校正决定系数和交叉验证决定系数分别是0.99582和0.98687,校正均方差和交叉验证均方差分别是0.1370和0.1900;过氧化值模型中,校正决定系数和交叉验证决定系数分别是0.99999和0.99926,校正均方差和交叉验证均方差分别是0.756×10-4和0.684×10-3。用该模型对验证集样本进行预测并统计分析,表明预测值与测定值无显著差异,傅里叶近红外光谱技术快速定量检测腊肉酸价和过氧化值是可行的。     

基于神经网络的生鲜水产品货架期预测模型的构建及应用

为建立一种能够同时适用于多种新鲜水产品货架期的预测模型,采用反向传播(BP)神经网络模型、遗传算法(GA)优化的BP神经网络模型(GA-BP)、径向基函数(RBF)神经网络模型、极限学习机(ELM)神经网络模型和支持向量回归机(SVR)模型分别对金枪鱼、三文鱼、大菱鲆和鲷鱼的货架期进行预测,寻找最优的模型预测结果。首先通过试验获得4种水产品在0,4,10 ℃贮藏条件下的感官评分、菌落总数、挥发性盐基氮值、K值、pH值,构建训练样本和测试样本。经相关性分析,选择与水产品货架期相关性较高的感官评分、菌落总数、挥发性盐基氮值、K值作为模型的输入层单元,然后确定各模型的网络拓扑结构以及参数,进行模型的训练,最后使用训练好的5种模型对测试样本的货架期进行预测。结果表明:5种预测模型的预测精度排序为:SVR模型>RBF神经网络模型>GA-BP神经网络模型>ELM神经网络模型>BP神经网络模型,其中BP神经网络模型的预测精度最差,均方误差(MSE)为9.5127×10-4,平均绝对误差(MAE)为0.0197,平均绝对百分比误差(MAPE)为0.0825,R2为0.9766;SVR模型的预测精度最优,预测误差均在12%以内,MSE为2.2971×10-4,MAE为0.0128,MAPE为0.0631,R2为0.9944,能够很好地同时预测4种水产品在不同储藏温度下的货架期。本研究为水产品的品质控制提供一定的理论基础。     

基于改进型RFB-MobileNetV3的棉杂图像检测

针对深度卷积神经网络模型复杂导致嵌入式设备难以满足实时在线检测的问题,提出改进型RFB-MobileNetV3(RFB-MNV3)的棉杂检测方法。首先,根据高精度轻量化网络模型的构建和保证检测准确率高的前提下,减少MNV3冗余网络结构;其次,将3×3卷积层取代5×5卷积层,1×3+3×1卷积层折叠取代3×3卷积层作为改进型感受野(RFB)模块部署到改进型MNV3的池化层中,以提升棉杂的在线检测速度和准确率;最后,将改进前后的算法与其它检测算法进行比较。结果表明,改进型RFB-MNV3算法的单张检测速度可达到0.02 s,在线检测平均准确率达到89.05%。通过对MNV3网络结构进行改进,在保证高检测准确率的前提下,可满足嵌入式设备在线棉杂检测的需求。     

北京市市售牡蛎中诺如病毒污染对居民健康影响的初步定量风险评估

目的通过建立针对北京市居民从零售到餐桌的牡蛎-诺如病毒暴露模型,对北京市居民通过市售牡蛎暴露诺如病毒的风险开展初步定量评估。方法利用北京市市售牡蛎中诺如病毒的356条定量检测数据,模拟北京市居民牡蛎消费情景,建立暴露评估模型,结合文献发表的基于暴发数据推导的剂量-反应关系模型,对居民通过生食牡蛎后发生诺如病毒感染的健康风险及其影响因素进行评估。结果 GI型和GII型诺如病毒阳性样品的污染水平分别为2.62×104个病毒拷贝/g(95%CI:3.73×103~1.54×105)和5.02×104个病毒拷贝/g(95%CI:8.13×103~2.52×105)。对血清分泌受体阳性的人群,生食1个可能受GI型和GII型诺如病毒污染的牡蛎的发病风险分别为0.93(95%CI:0.73~0.98)和0.95(95%CI:0.80~0.99);对血清分泌受体阴性的人群,生食1个可能污染GI型和GII型诺如病毒牡蛎的发病风险分别为0.37(95%CI:0.04~0.64)和0.57(95%CI:0.07~0.99)。风险的估计值与诺如病毒阴性样品的赋值相关,相关系数为0.49。不确定性分析显示,现有检测方法的检出限高于半数感染及致病剂量,是评估结果不确定性的主要来源。结论北京市居民通过生食牡蛎发生食源性诺如病毒感染的风险很高;提高病毒拷贝定量检测技术的灵敏度,降低检测方法的检出限水平是降低评估结果不确定性的主要途径。     

Impedimetric Determinaton of Total, Mesophilic and Psychrotrophic Zounts in Raw Milk

A rapid impedimetric determination for total, mesophilic and psychrotrophic counts in raw milk showed correlations (between impedance detection times and standard plate counts) of -0.96, -0.95, and 0.96, respectively. Mesophiles were most often seen as the predominant population, the impedimetric method allowed for these samples containing above 1.0 × 10⁵ CFU/ml to be screened out within 4 hr. Psychrotrophic levels of 1 × 10⁵ CFU/ml and above were screened within 21 hr, while total concentration of samples containing above 1 × 10⁵ CFU/ml were screened within 16 hr.     

A comparison of contents of group A and B trichothecenes and microbial counts in different cereal species

Concentrations of trichothecenes and the amounts of microbial biomass were compared in grain of bread wheat, durum wheat, triticale, rye, oat and barley. Grain samples came from lines regionalised in Poland grown under identical climatic and agricultural conditions in 2007. Among the six analysed cereals, the highest mean concentration of toxic metabolites of 151.89?µg/kg was found for grain of Triticum durum, whereas the lowest was for barley grain (25.56?µg/kg). The highest contamination with microscopic fungi was recorded in case of barley and rice grain (mean concentrations of ergosterol were 12.53?mg/kg and 11.24?mg/kg, respectively). In case of the analysed cereals, the total microbial biomass expressed in the amount of ATP turned out to be the highest (4.7?×?10⁵ relative light units [RLU]) for rye and oat (3.2?×?10⁵ RLU). The results of the applied classical discrimination analysis indicate a significant diversification of species in terms of all the 11 analysed metabolites.     

Artificial neural network modeling of mass transfer during osmotic dehydration of kaffir lime peel

Mass transfer of kaffir lime peel during osmotic dehydration was investigated in this paper. Processing factors were solute concentrations, process temperatures, and immersion time. The results showed that increasing solute concentration and process temperature resulted in a higher reduction in moisture contents of kaffir lime peel and increase in water loss and solid gain rates. Analysis of variance showed significant effects (P < 0.05) of all processing factors except process temperatures for water loss. Multilayer feedforward neural network (MFNN) was proposed to predict percentages of water loss and solid gain of kaffir lime peel during osmotic dehydration based on three processing factors as inputs. The best network with the lowest average mean squared error (MSE) of 0.0066 and the highest average regression coefficient (r²) of 0.9725 from normalized training and validating data sets was composed of one hidden layer with five hidden neurons and used Levenberg–Marquardt algorithm as a training algorithm. A simulation test showed good generalization of the successfully trained MFNN model with the average MSE of 6.5813 and 5.9340, and average r² of 0.9745 and 0.9632, respectively, for water loss and solid gain. Compared with multiple linear regression models, MFNN was found to be more suitable for predicting water loss and solid gain during the OD process of kaffir lime peel.     

Physiological Activity of Deep-Frozen Concentrates of Leuconostoc Strains

11 strains of Leuconostoc cremoris and 1 strain of Leuconostoc lactis were the subject of investigations. Basing on the evaluation of growth parameters in the different cultivation media and on the diacetyl production ability in milk 2 strains of Leuconostoc cremoris and 1 strain of Leuconostoc lactis were chosen for preparing frozen concentrated biomass. Sterile milk or cream (18% fat) were used as protecting agents. Concentrated biomass was frozen at ?70 °C and stored at ?30 °C for 3 months. The prepared concentrates contained from 1.6 × 10¹¹ to 2.8 × 10¹¹ colony forming units (CFU)/g. Except for one (strain “S”) the strains retained 100% survival. All the strains showed high growth activity after their recovery in milk (3.3 × 10⁸ - 8.4 × 10⁸ CFU/ml). Both protecting agents applied proved to be equally good for preserving viability of the cells and for their acidifying activity. In spite of the protecting agent a decrease in the aroma activity by 25–50% was observed after one month of the biomass storage. Nevertheless, concentrated frozen multicomponent starters containing Leuconostoc strains retained the ability to produce diacetyl and acetoin on the unchanged level during the three months storage period.     

Influence of ultrasound on mass transport during cheese brining

 The effect of ultrasound on mass transfer during cheese brining has been investigated. The rate of water removal and NaCl gain increased when ultrasound was applied in comparison with brining performed under static or dynamic conditions, suggesting that ultrasound improves both external and internal mass transfer. A simple diffusional model was developed to simulate mass transport during acoustic brining. Model parameters were estimated using experimental data from acoustic brining experiments carried out on cheese cylinders of 1.7×10^–2m radius and 3×10^–2 m height at different temperatures (5, 15 and 20  °C). Effective water (D _W) and NaCl (D _S) diffusivities estimated using the proposed model ranged from 5.0×10^–10m²/s and 8.0×10^–10 m²/s at 5  °C to 1.3×10^–9m²/s and 1.2×10^–9 m²/s at 20  °C. Both D _W and D _S varied with temperature according to the Arrhenius equation. Through the proposed model, water losses and NaCl gains of the experiments used in the parameter identification were accurately simulated (average %var=98.2%) and also of two additional acoustic experiments carried out under different conditions of temperature (10  °C) and sample size and geometry [parallelepiped of (6×2.5×1.25)×10^–2 m] to those used in the parameter identification (average %var=98.4%). Received: 22 September 1998 / Revised version: 20 November 1998     

Gamma-Irradiation Affects Some Physical Properties of Dry Bean (Phaseolus vulgaris) Starch

Gamma-irradiated great northern (GN) bean starch (2.5–20 kGy) showed apparent increased susceptibility to central fissures. Bean damage during scanning electron microscopy was observed using ≥400OX magnification on GN bean starch samples. Approximate molecular weights estimated by gel filtration chromatography Sepharose Cl-2B of control GN bean starch were: amylopectin >2×10⁶ and amylose 2×10⁵. Amylose-like fraction of irradiated bean starch (20 kGy) revealed two peaks ～6.9×10⁴ and 1.5×10⁵ daltons. Differential scanning calorimetry showed an increase in gelatinization enthalpy and a small increase in gelatinization peak temperature at 20 kGy suggesting a reorganization of the crystalline and amorphous phases of the starch granule.     

Aciduric and Heat Resistant Microorganisms in Apple Juice and Cider Processing Operations

Site visits to 15 apple juice or cider facilities were conducted in New York State. Washed apples, pulp, press juice, final product, and other samples, when appropriate, were sampled aseptically. Samples were plated on acidified potato dextrose agar, before and after heating at 70°C for 1-2 hr. For unheated samples, molds predominated in fruit, yeasts and molds in pulp, and yeasts in press and final juice samples. Average counts were 2.8 × 10⁴, 7.3 × 10⁴, 1.7 × 10⁵, and 1.4 × 10⁵/g, respectively. Heat resistant yeasts, molds, and bacteria were isolated frequently, generally at a level of less than 1 per log.     

Diffusion of lactic acid in a buffered gel system supporting growth of Lactobacillus curvatus

Diffusion of lactic acid in a buffered gel system was investigated using fluorescent ratio imaging. The range of pH studied was from 5.5 to 3.8, which included the minimum pH of growth of Lactobacillus curvatus (4.26). Diffusion of lactic acid in this pH range was found to be Fickian and the estimated effective diffusion coefficient of lactic acid at 20 °C was 2.81 × 10^?10 m² s^?1 with a standard deviation of 0.21 × 10^?10 m² s^?1. This effective diffusion coefficient was estimated by fitting the one‐dimensional solution of the Fickian diffusion equation for infinite systems to a transformed experimental data set. The original data set as recorded by the fluorescent microscope consisted of concentration–distance curves at specific time intervals. This was transformed by realigning the distance scale of all the concentration–distance curves. © 2002 Society of Chemical Industry     

The effect of pitching rate on fermentation,maturation and flavour compounds of beer produced on an industrial scale

The aim of the study was to determine the effect of the initial number of yeast cells in the wort on the process of fermentation, maturation and the content of the volatile components of beer, as well as the viability and vitality of the yeast biomass. The experiments were performed on an industrial scale, with fermentation and maturation in cylindro‐conical fermentation tanks with a capacity of 3800 hL. Yeast for pitching was collected after secondary fermentation (third passage) and wort pitching levels were 5 × 10⁶, 7 × 10⁶ and 9 × 10⁶ cells/mL. During fermentation and maturation, the changes in the content of the extract, yeast growth, yeast vitality and selected volatile components were investigated. Experiments showed that the yeast inoculum had a significant impact on the course of the fermentation and metabolic changes. With increasing numbers of cells introduced into the wort, the content of the esters and fusel alcohols increased, while the acetaldehyde concentration decreased. These changes affected the final quality of the beer. Copyright © 2015 The Institute of Brewing & Distilling