pH对红色红曲菌（Monascus ruber）M7生长及产孢的影响

研究红色红曲菌（Monascus ruber）M7生长、繁殖两个不同阶段对环境pH需求的差异,在孢子培养过程中对pH进行分段控制,旨在减小产孢条件对菌丝生长的影响,提高孢子产量。结果表明,菌株M7孢子在pH 3.0～6.0条件下培养8 h均可100%萌发,其中,pH 4.0～5.0最佳,萌发最为迅速,在pH 7.0～8.0时孢子萌发受到严重抑制,萌发率很低;菌落生长在pH 4.0～6.0时较快,其中,pH 5.0最佳,菌落直径最大,但在pH 3.0和7.0时菌落生长明显减慢,在pH 8.0时受到严重抑制,形成的菌落很小;pH＞8.0的碱性条件为极端pH条件,孢子萌发和菌落生长完全被抑制。在菌株M7生长的pH范围内,产孢能力随着pH的升高而提高,适宜的产孢pH范围为7.0～8.0。根据菌株M7生长和繁殖对pH的需求差异,控制培养前期pH值为5.0、后期pH值为7.0～8.0,在PDB和CYB培养基中的产孢量分别提高了12～16倍和3～4倍。     

基于径向基函数神经网络探索风味物质对啤酒风味的预测

以啤酒中对风味贡献较大的高级醇、酯、有机酸、酒精、苦味质共13种风味物质作为输入指标,以感官品评评分为输出指标,利用径向基函数(Radial basis function,RBF)神经网络(neural network)对国内外20种啤酒的风味进行预测,准确率达75%,与其他相同样本数的神经网络相比较,准确率有较大提高。说明啤酒中这13种风味物质与综合风味之间存在密切关系,以利于啤酒生产厂家对啤酒整体风味的控制。     

Nondestructive determination of salmon fillet freshness during storage at different temperatures by electronic nose system combined with radial basis function neural networks

This study develops a predictive model for determining freshness of salmon fillets during cold storage at different temperatures using electronic nose combined with principal component analysis (PCA) and radial basis function neural networks (RBFNNs). The electronic nose sensed ammonia/amines, hydrocarbons, solvents and aromatics that increased during storage. The concentrations of the volatiles were compared with the increased thiobarbituric acid (TBA), total volatile basic nitrogen (TVB-N), total aerobic bacteria count (TAC) and decreased of sensory assessments (SA). Gas chromatograph–ion mobility spectrometry analysis confirmed the changes in gas species. RBFNNs and PCA were used to establish predictive models and the relative errors of TBA, TVB-N and TAC by the PCA-RBFNNs model were all within ±10% and SA was within ±15%. These results suggest that the PCA-RBFNNs model can be used to predict changes in the freshness of salmon fillets stored at −2 to 10 °C.     

短裤特征截面曲线的径向基函数神经网络模型构建

将着装人台进行三维扫描获取点云数据,截取与人体特征部位相对应的短裤特征截面。将原数据坐标点转化为极坐标系下的极角与极径值后,以极角值作为输入向量,极径值作为输出向量,构建短裤特征截面曲线的径向基函数（RBF）神经网络模型,并与反向传播（BP）神经网络、最小二乘法及三次样条函数的拟合效果进行比较。结果表明,神经网络拟合曲线的平均绝对误差比最小二乘法及三次样条函数方法小,仿真输出曲线和原始数据非常接近,且曲线光滑;RBF网络的训练速度更快,所需训练步数少,拟合效率明显优于BP神经网络。     

Optimizing a submerged Monascus cultivation for production of red pigment with bug damaged wheat using artificial neural networks

The combined effect of temperature, agitation speed, and light on red pigment production by Monascus purpureus (M. purpureus) Went DSM 1604 using bug damaged wheat was studied using an artificial neural network (ANN). Information retrieved from the ANN was used to determine the optimal operating conditions for pigment production by M. purpureus using bug damaged wheat meal. The developed ANN had R ² values for training, validation, and testing data sets of 0.993, 0.961, and 0.944, respectively. According to the model, the highest pigment production of 1.874 absorbance units at 510 nm (A_{510 nm}) would be achieved at 29°C and 150 rpm under light conditions. The mean value of the experimental results obtained under these optimum conditions was 1.787±0.072 A_{510 nm}, corresponding to a pigment yield of 35.740 A_{510 nm}/g. The study showed that bug damaged wheat can be used as a substrate for red pigment production by M. purpureus.     

Predicting torque of worsted singles yarn using an efficient radial basis function network-based method

The torque in single-spun yarns is an inherent property of the twisting and bending of staple fibres during the formation of yarn combined with the effect of applied tension on the yarn. The consequences of yarn torque are well known and are widely observed as yarn instability, e.g., yarn rotation under tension; local snarling and entanglement at low loads, and as distortion in fabric, i.e., edge-curl and skewing in knitted fabric. In this paper, a method for predicting the yarn torque based on the radial basis function networks is presented and evaluated. This method uses a “universal approximator” based on neural network methodology to minimize noise during training of the network and to approximate the yarn torque as a function of the geometrical and physical parameters of yarns (twist, linear density) and the applied load. The current method is an integral radial basis function network-based approach suitable for textile engineering and gives very good prediction of yarn torque across a range of yarn structural parameters and test conditions.     

A radial basis function neural network approach to determine the survival of Listeria monocytogenes in Katiki, a traditional Greek soft cheese

A radial basis function neural network was developed to determine the kinetic behavior of Listeria monocytogenes in Katiki, a traditional white acid-curd soft spreadable cheese. The applicability of the neural network approach was compared with the reparameterized Gompertz, the modified Weibull, and the Geeraerd primary models. Model performance was assessed with the root mean square error of the residuals of the model (RMSE), the regression coefficient (R2), and the F test. Commercially prepared cheese samples were artificially inoculated with a five-strain cocktail of L. monocytogenes, with an initial concentration of 10(6) CFU g(-1) and stored at 5, 10, 15, and 20 degrees C for 40 days. At each storage temperature, a pathogen viability loss profile was evident and included a shoulder, a log-linear phase, and a tailing phase. The developed neural network described the survival of L. monocytogenes equally well or slightly better than did the three primary models. The performance indices for the training subset of the network were R2 = 0.993 and RMSE = 0.214. The relevant mean values for all storage temperatures were R2 = 0.981, 0.986, and 0.985 and RMSE = 0.344, 0.256, and 0.262 for the reparameterized Gompertz, modified Weibull, and Geeraerd models, respectively. The results of the F test indicated that none of the primary models were able to describe accurately the survival of the pathogen at 5 degrees C, whereas with the neural network all fvalues were significant. The neural network and primary models all were validated under constant temperature storage conditions (12 and 17 degrees C). First or second order polynomial models were used to relate the inactivation parameters to temperature, whereas the neural network was used a one-step modeling approach. Comparison of the prediction capability was based on bias and accuracy factors and on the goodness-of-fit index. The prediction performance of the neural network approach was equal to that of the primary models at both validation temperatures. The results of this work could increase the knowledge basis for the applicability of neural networks as an alternative tool in predictive microbiology.     

Modelling the growth dynamics of interacting mixed cultures: a case of amensalism

In this work, a specific membrane bioreactor was used to perform co-cultures of two Saccharomyces cerevisiae yeast strains: a killer strain and a sensitive strain. Biomass could be segregated into four groups: viable killer yeasts, dead killer yeasts, viable sensitive yeasts and dead sensitive yeasts. An existing mathematical model describing the population dynamics in the mixed killer/sensitive cultures was confronted with the new experimental data. As it gave poor accuracy, some improvements were proposed and tested. In particular, a lag phase before the beginning of the lethal interaction between the two strains was introduced, in correspondence to the experimental observations.     

Comparison between the Arrhenius model and the radial basis function neural network (RBFNN) model for predicting quality changes of frozen shrimp (Solenocera melantho)

Changes in quality indices [total volatile base nitrogen (TVB-N), salt extractable protein (SEP), hypoxanthine (Hx), K-value, sensory assessment (SA), and electrical conductivity (EC)] for shrimp (Solenocera melantho) stored at ?28, ?20, and ?12°C for 112 days were investigated in this study. The Arrhenius model and the radial basis function neural network (RBFNN) model were established to predict changes in the quality of shrimp during storage. Quality of shrimp stored at ?12°C changed more quickly during 56–112 days, but those stored at ?28°C deteriorated slowly during the entire storage period. Additionally, the indicators SEP, EC, and SA all fitted to the Arrhenius model well (relative errors within ±10%), but this model did not perform well in the prediction of K-value, Hx, and TVB-N on some days. However, the RBFNN model showed excellent accuracy for all indicators (relative errors within ±0.5%). The RBFNN model performed better than the Arrhenius model in predicting the quality of shrimp stored at ?28°C to ?12°C.     

Verification of the geological origin of bottled mineral water using artificial neural networks

As a first step towards objective and cost-efficient verification of the geographical origin of commercially sold mineral water, we determined up to what extent the chemical composition of mineral water can be linked to the geology of the local water source. For this purpose, a dataset consisting of 145 European mineral water samples from a known geology was analysed using counter-propagation artificial neural networks (CP-ANNs) with supervised learning algorithm. The models were tested for recall ability (RA) and validated with a leave-one-out cross validation (LOO-CV).     

配毛中原料品质的BP神经网络预测研究

用BP神经网络方法对配毛时所需的羊毛原料品质进行预测，阐述了利用BP神经网络预测羊毛品质的工作原理，给出了羊毛品质预测系统的BP神经网络模型，并得出了试验结论。     

应用人工神经网络技术建立航空配餐微生物生长模型

首次将预测微生物学应用于航空配餐,利用人工神经网络(ANN)技术建立配餐微生物数学模型。选用猪肉冷荤作为研究食品,大肠杆菌、金黄色葡萄球菌和猪肉冷荤本底菌群作为研究微生物;猪肉冷荤灭菌后染有不同浓度组合的研究微生物,检测不同氯化钠和温度水平下,储存2～27h微生物生长繁殖结果:应用人工神经网络技术对大肠杆菌、金黄色葡萄球菌和本底菌群分别建立生长模型,以及3种微生物合并建立模型;通过筛选确定以3种微生物合并建模的第2号MLP模型作为最佳模型,网络拓扑结构为6∶5∶3,训练准确率、测试准确率和验证准确率分别为0.848329、0.871849和0.897135。人工神经网络建立的数学模型能较好地应用于航空配餐冷荤多种微生物生长的预测。     

Application of neural networks to simulate the growth profile of lactic acid bacteria in green olive fermentation

The growth profile of five strains of lactic acid bacteria (Lactobacillus plantarum ACA-DC 287, L. plantarum ACA-DC 146, Lactobacillus paracasei ACA-DC 4037, Lactobacillus sakei LQC 1378, and Leuconostoc mesenteroides LQC 1398) was investigated in controlled fermentation of cv. Conservolea green olives with a multilayer perceptron network, a combined logistic-Fermi function, and a two-term Gompertz function. Neural network training was based on the steepest-descent gradient learning algorithm. Model performance was compared with the experimental data with five statistical indices, namely coefficient of determination (R2), root mean square error (RMSE), mean relative percentage error (MRPE), mean absolute percentage error (MAPE), and standard error of prediction (SEP). The experimental data set consisted of 125 counts (CFU per milliliter) of lactic acid bacteria during the green olive fermentation process for up to 38 days (5 strains x 25 sampling days). For model development, a standard methodology was followed, dividing the data set into training (120 data) and validation (25 data) subsets. Our results demonstrated that the developed network was able to model the growth and survival profile of all the strains of lactic acid bacteria during fermentation equally well with the statistical models. The performance indices for the training subset of the multilayer perceptron network were R2 = 0.987, RMSE = 0.097, MRPE = 0.069, MAPE = 0.933, and SEP = 1.316. The relevant mean values for the logistic-Fermi and two-term Gompertz functions were R2 = 0.981 and 0.989, RMSE = 0.109 and 0.083, MRPE = 0.026 and 0.030, MAPE = 1.430 and 1.076, and SEP = 1.490 and 1.127, respectively. For the validation subset, the network also gave good predictions (R2 = 0.968, RMSE = 0.149, MRPE = 0.100, MAPE = 1.411, and SEP = 2.009).     

Predicting survival of Escherichia coli O157:H7 in dry fermented sausage using artificial neural networks

The Canadian Food Inspection Agency required the meat industry to ensure Escherichia coli O157:H7 does not survive (experiences > or = 5 log CFU/g reduction) in dry fermented sausage (salami) during processing after a series of foodborne illness outbreaks resulting from this pathogenic bacterium occurred. The industry is in need of an effective technique like predictive modeling for estimating bacterial viability, because traditional microbiological enumeration is a time-consuming and laborious method. The accuracy and speed of artificial neural networks (ANNs) for this purpose is an attractive alternative (developed from predictive microbiology), especially for on-line processing in industry. Data from a study of interactive effects of different levels of pH, water activity, and the concentrations of allyl isothiocyanate at various times during sausage manufacture in reducing numbers of E. coli O157:H7 were collected. Data were used to develop predictive models using a general regression neural network (GRNN), a form of ANN, and a statistical linear polynomial regression technique. Both models were compared for their predictive error, using various statistical indices. GRNN predictions for training and test data sets had less serious errors when compared with the statistical model predictions. GRNN models were better and slightly better for training and test sets, respectively, than was the statistical model. Also, GRNN accurately predicted the level of allyl isothiocyanate required, ensuring a 5-log reduction, when an appropriate production set was created by interpolation. Because they are simple to generate, fast, and accurate, ANN models may be of value for industrial use in dry fermented sausage manufacture to reduce the hazard associated with E. coli O157:H7 in fresh beef and permit production of consistently safe products from this raw material.     

Classification of dry-cured hams according to the maturation time using near infrared spectra and artificial neural networks

An attempt to classify dry-cured hams according to the maturation time on the basis of near infrared (NIR) spectra was studied. The study comprised 128 samples of biceps femoris (BF) muscle from dry-cured hams matured for 10 (n = 32), 12 (n = 32), 14 (n = 32) or 16 months (n = 32). Samples were minced and scanned in the wavelength range from 400 to 2500 nm using spectrometer NIR System model 6500 (Silver Spring, MD, USA). Spectral data were used for i) splitting of samples into the training and test set using 2D Kohonen artificial neural networks (ANN) and for ii) construction of classification models using counter-propagation ANN (CP-ANN). Different models were tested, and the one selected was based on the lowest percentage of misclassified test samples (external validation). Overall correctness of the classification was 79.7%, which demonstrates practical relevance of using NIR spectroscopy and ANN for dry-cured ham processing control.     

Predicting the strength reduction of particleboard subjected to various climatic conditions in Japan using artificial neural networks

Particleboard specimens were subjected to various climatic conditions in Japan, and the relationships between climatic factors and internal bond strength (IB) were investigated using multiple regression analysis (MRA) or artificial neural networks (ANN). At low- and middle-temperature sites, the IB predicted using MRA (IB_MRA) and ANN (IB_ANN) decreased linearly with increasing exposure time. In addition, at high-temperature sites, with increasing exposure time, IB_MRA decreased linearly, whereas IB_ANN decreased exponentially. The trend of IB_ANN was almost identical to that of the measured IB of the specimens subjected to various climatic conditions. Moreover, IB_MRA and IB_ANN for 1-, 3-, and 5-year exposures were predicted using nationwide climatic factors. The minimum IB is zero when the particleboard is deteriorated; however, negative IB was predicted using MRA when the exposure time increased in the high-temperature area. In addition, the IB for 1-year exposure in the low-temperature area near site 1 was higher than the initial IB of 0.833 MPa. MRA is not always valid because of the assumption of linearity. However, negative IB even for 5-year exposure in the high-temperature area and high IB even for 1-year exposure in the low-temperature area were not predicted using ANN. The IB reduction was predicted correctly using ANN, and the correct IB reduction could be mapped.     

Modelling of the individual and combined effects of water activity and temperature on the radial growth of Aspergillus flavus and A. parasiticus on corn

A full factorial design of five temperatures (16, 22, 25, 30 and 37 degrees C) and seven a(w) values between 0.801 and 0.982 was used to investigate the growth of the two major aflatoxin producing Aspergillus isolates on corn. The colony growth rates (g, mmd(-1)) and lag phases (lambda, d) were estimated by fitting a flexible primary growth model. Subsequently, secondary models relating g or lambda to a(w) or temperature or a(w) and temperature combined, were developed and validated by using independently collected data. The Gibson and linear Arrhenius-Davey model describing the individual effects of a(w) or temperature on g or lambda proved an adequate predictor of either growth parameter. Based on the validation criteria, a quadratic polynomial function proved to be more suitable than a Gaussian function or extended Davey model for describing the combined effect of a(w) and temperature on g or lambda. Both isolates studied had optimum growth temperatures of approximately 30 degrees C. No growth was observed for both isolates at a(w) 0.801, growth only occurring at 25 and 30 degrees C at a(w) 0.822. Significant interaction between a(w) and temperature on g and lambda was observed for both isolates. The developed models can be applied in the preservation of corn and the development of models that incorporate other factors important to mould growth on corn.     

Predicting fungal growth: the effect of water activity on Penicillium roqueforti.

The effect of water activity on the colony growth of Penicillium roqueforti is studied by predictive modelling techniques. Measured colony diameter growth curves are fitted to estimate the growth rate and lag phase of the curves. The colony growth rate was modelled by a quadratic function of transformed water activity (a(w)) values, as suggested by Baranyi et al. (Food Microbiol. 10 (1993) 43-59). The lag time was modelled as a function of water activity, by means of the sum of a constant term and a hyperboloid function of a(w) raised to the second power. The lag-phase of Penicillium roqueforti was found insensitive to the water activity in the range of its higher (a(w) > 0.92) values.     

Modelling of solid-phase tea waste extraction for the removal of manganese from food samples by using artificial neural network approach

In this study, a three-layer artificial neural network (ANN) model was employed to develop prediction model for removal of manganese from food samples using tea waste as a low cost adsorbent. After removal of manganese from food samples with acetic acid (5 mol L⁻¹), manganese was adsorbed to a small amount of tea waste, desorbed with nitric acid as a eluent solvent, and determined by flame atomic absorption spectrometry. The input parameters chosen of the model was pH, amount of tea waste, extraction time and eluent concentration. After backpropagation (BP) training, the ANN model was able to predict extraction efficiency of manganese with a tangent sigmoid transfer function at hidden layer and a linear transfer function at output layer. Under the optimum conditions, the detection limit was 0.6 ng g⁻¹. The method was applied to the separation, pre-concentration and determination of manganese in food samples and one reference material.