جستجوی مقالات مرتبط با کلیدواژه "شبکه ی عصبی مصنوعی" در نشریات گروه "صنایع غذایی"

Nowadays, citrus fruits, especially oranges, is very important in the human nutrition regime, and its quality characteristics assessment is very important. This study aimed to predict some biochemical characteristics of blood orange, using machine vision and artificial neural networks. In this experiment, the amount of vitamin C content, sugar content, and acidity (pH) were obtained using destructive laboratory methods. Images of blood orange samples were captured and 108 texture features and 57 color features were extracted on CIElab, RGB, HSV, and HIS color spaces and finally, the artificial neural networks method has been used to estimate the desired properties. To evaluate and select the most optimal artificial neural network, a feed-forward neural networks with Levenberg-Marquardt learning algorithm, the different number of neurons, and different transfer functions in the hidden and output layers was used. Finally, using the best neural network and 165 textural-color features, the amount of vitamin C content, sugar content, and pH were estimated with a correlation coefficient of 0.950, 0.968, and 0.884, respectively. Therefore, considering the appropriate correlation coefficient, machine vision and image processing technology can estimate some biochemical characteristics of blood oranges accurately.

The aim of this research was to investigate the degradation kinetics of the major quality properties of medlar (Mespilus germanica) during cold storage. Medlar is a widely growth in northern Iran and its fruit is used as a nutritional component and as a medicinal remedy. In fruits, quality properties are used as a consumer-based criteria of acceptability. So it is important to evaluate parameters that affected the medlar quality. Measurement of these parameters is an expensive and time-consuming process. Therefore, parameter prediction due to affecting factors will be more useful. In the present research, mathematical models and artificial neural networks (ANN) were used for modelling the relationship between physicochemical properties and color attributes with cold storage time. Five kinetic models viz. zero order, first order, Second order, fractional conversion and Weibull models were used for modelling using MATLAB. Among the kinetics models, the Weibull model was found to be more suitable to predict the changes in all physicochemical ( , ) and color ( , ) parameters. In ANN, multi-layer perception (MLP) used with different number of neurons. The network’s inputs include storage time, medlar moisture content and ripening stage and the network’s output were the values of the physicochemical and color properties. The training rule was Momentum Levenberg-Marquardt. The transfer functions were Tansig, Purelin and Logsig. The results showed that MLP network with Levenberg-Marquardt training function, Purelin transfer function and 3-8-4-3 and 3-7-2 topologies had the best accuracy for prediction of for physicochemical and color properties. This network can predict physicochemical and color properties of the medlar with  coefficient of 0.9983 and 0.9992 and MSE of 0.021, 0.000008 and 0.000059 respectively.

The aim of this study was to determine the kinetics, effective moisture diffusivity, activation energy, specific energy consumption, and also predict the moisture content of white mulberry during the drying process with microwave-hot air dryer using mathematical models, Artificial Neural Networks (ANN) and Neuro-Fuzzy Inference System (ANFIS). Drying process was accomplished in three temperature levels (40, 55, and 70°C), three inlet air velocity levels (0.5, 1 and 1.5 m/s) and three microwave power levels (270, 450 and 630 W). To estimate the moisture ratio of white mulberry, 10 mathematical models, ANN and ANFIS were used to fit the experimental data of thin-layer drying. The results showed, the maximum and minimum effective moisture diffusivity during drying was calculated 3.56×10-9 and 3.86×10-10 m2/s, respectively. Also, the minimum and maximum effective moisture diffusivity during drying was achieved 48.54 and 1380.88 Mj/kg, respectively. Among the mathematical models under study, the Page model was the best model for describing the behavior of the thin layer of white mulberry drying. The mean square error (MSE) values for the mathematical models, ANN, and ANFIS were 0.00059, 0.0052 and 0.0044, respectively. Therefore, the ANFIS model with the highest Correlation Coefficient (R2=0.99995), the least percentage of mean relative error (ε=1.84) and mean square error (MSE=0.0044) were used to evaluate the moisture ratio in comparison with other methods implemented in this research Selected as the best model

In this research to model the process of extracting oil from Niger seeds using the combinational pretreatment of microwave-pulsed electric field three microwave time levels of (0, 100, and 200 seconds) and three electric field intensity levels of (0, 2.5 and 5 kV/cm) were used and after applying these pretreatments, the oil of seeds were extracted using the screw press with different speeds (11 to 57 rpm) and the efficiency of oil extraction process, density, color, oxidative stability, phenolic compounds and protein amount of meal were considered. The artificial neural network tool was used to predict the variations process. Through studying and examining various networks, the feed forward back propagation network with 6-9-3 topologies and with correlation coefficient of more than 0.995 and mean squared error less than 0.001 using logarithm sigmoid activation function, resilient learning pattern and learning process of 1000 were determined as the best neural method. On the other hand the results indicated that an increase in the microwave time and also in the electric field intensity at first led to increase in the efficiency of oil extraction process but with more increase in these two parameters the efficiency amount of oil extraction process was decreased. Also with increase in the microwave time, electric field intensity and the rotational speed of screw press the acidity amount of oils was increased too. The results obtained from selected optimized models were evaluated too and these models with high correlation coefficient (over 0.918) were able to predict the variation process of oil samples produced using microwave-pulsed electric field pretreatment.

In this research work, in order to estimate the drying properties of eggplant and turnip in a combined microwave-convection dryer it was used artificial neural network method in order to estimate the Drying process was accomplished in three temperature levels (40, 55, and 70 °C), three inlet air velocity levels (0.5, 1.1 and 1.7 m/s) and three microwave power levels (270, 450 and 630 W) in a combined microwave-convection dryer that the three parameters were utilized as input in predicting the effective moisture diffusion coefficient and specific energy consumption in artificial neural network. A feed forward and cascade forward back propagation neural network with training functions of Levenberg - Marquardt (LM) and Bayesian Regulation (BR) for training of patterns. According to results, the highest value of the effective moisture diffusion coefficient for eggplant and turnip was obtained 3.39×10-9 and 3.05×10-9 m2/s, respectively. Results of ANN investigations showed that the optimum feed forward back propagation network with 3-20-20-2 topology and training function of Levenberg – Marquardt could predict the effective moisture diffusion coefficient and specific energy consumption with determination coefficients of 0.9821 and 0.9952 and mean square error of 0.00014 in various drying conditions of eggplant and turnip. Also the highest determination coefficients for prediction of drying rate and moisture ration obtained 0.9698 and 0.9988 with the value of mean square error of 0.0045 in cascade back propagation neural network with training algorithm of Levenberg – Marquardt.

Cucumber fruit consumes in high quantities during all seasons in Iran; so it is important to evaluate parameters that affected the cucumber quality. Measurement of these parameters is expensive and time-consuming process. Therefore, parameters prediction due to affecting factors will be more useful. In this research work, artificial neural networks used for modelling the relationship between mechanical properties (crispness index and firmness) and sound pressure during cutting (acoustic response) with storage time, storage conditions on different positions on cucumber fruit. The networks input were storage time, storage conditions and test position on the fruit length. The networks output targets were the values of the mechanical properties and acoustic response. The different networks defined and trianed with different topologies. Multi layer perception (MLP) and radial basis function (RBF) networks used with different number of neurons. The training rules were Momentum, Conjugate Gradient and Levenberg-Marquardt. The transfer functions were TanhAxon and SigmoidAxon. The networks evaluated respected to estimation the accuracy of acoustic and mechanical properties. The results showed that MLP network with Momentom training function, TanhAxon transfer function and 3-5-3-3 topology had the best accuracy for prediction of acoustic and mechanical properties of Viola cucumber fruit. This network can predict the sound pressure, crispness index and firmness of the fruit with determinations coefficient of 0.9973, 0.9456 and 0.9129 and root mean square error of 0.021, 0.052 and 0.059 respectively.

In this study, Response Surface Methodology (RSM) was used to optimize osmo-dehydration of orange slice. Effect of osmotic solution temperature in the range of 30 to 60 °C, immersion time from 0 to 300 min and sucrose concentration from 35 to 65 brix degree on water loss, solid gain, moisture content, water loss to solid gain ratio and brix change were investigated by Central Composite Design (CCD). Applying response surface and contour plots optimum for osmotic dehydration were found to be at temperature of 30 °C, immersion time of 229.2 minute and sucrose concentration of 65%. At this optimum point, water loss, solid gain, WL/SG ratio, moisture content (dry base) and brix difference were found to be 30.316 (g/100 g initial sample), 13.51 (g/100 g initial sample), 2.45, 2.77 % and 15.79, respectively. The result of artificial neural network indicated that the perceptron neural network with one hidden layer is able to anticipate the dehydration characteristics. This network predicted solid gain and moisture content with 5 neuron per hidden layers with R2 values of 0.937 and 0.959, respectively and brix difference and water loss with 30 neuron per hidden layer with R2 values of 0.961 and 0.942, respectively.

In this research, zucchini samples were soaked into different osmotic solution concentrations (sucrose 35%+sorbitol 5%, sucrose 35%+sorbitol 10% and sucrose 35%+sorbitol 15%) and at 5, 25 and 50℃ under ultrasound and non-ultrasound pertreatment. Water loss and solid gain were increased whereas moisture content was decreased for all samples during dehydration time. Water loss values were 35.75 and 31 (g/100g dry solid) at 5% sorbitol concentration and 40 and 35.42 (g/100g dry solid) for both samples of ultrasound and non-ultrasound treatment at 15% sorbitol concentration, respectively. These values indicated significant difference at α=1%. Furthermore, solid gain values were 9.33 and 5.08 (g/100g dry solid) for ultrasonicated and noultrasonicated samples at 5% sorbitol concentration and 12.58 and 6.17 (g/100g dry solid) at 15% sorbitol concentration, correspondingly.

In this research, thin-layer drying of tomato slices was simulated via a laboratory scale hot air dryer. The drying process was carried out at two different temperatures (60 and 70 oC). Drying data were fitted to 8 drying kinetic models. The goodness of fit was determined by means of R2, RMSE and χ2, it was concluded that Logarithmic model represented the best performance. Also, physicochemical properties of tomato slice such as shrinkage, acidity, pH and color were determined during drying. The result indicated that drying air temperature had a significantly affect on the color change of tomato slices. Furthermore, in this research, tomato slice favorite is evaluated by sensory evaluation. In this test sense qualities as color, aroma, flavor, appearance and chew ability (tissue brittleness) are considered. The results revealed that air temperature had a significant effect on the appearance (P<0.01). The results showed that perceptron neural network with logsig activation function as a goodness activation function can be estimated moisture ratio with R2 value 0.996.

The mechanical losses which occur in agricultural products are the damage''s which imposed on country economy. It is important to investigate the bruising phenomena، as an index of mechanical losses for loss reduction and optimization of harvest and postharvest machinery. In the current study، by means of an impact pendulum apparatus and by conducting a series of impact tests، the effects of temperature (0، 10، 20 and 300C)، variety (Golden Delicious and Red Delicious)، padding surface (corrugated carton، rubber and galvanized iron) and kinetic energy (300، 600 and 900 mJ) were investigated on rate of apple bruise. Statistical results showed that the effect of temperature، variety، padding surface and impact energy were significant on the mean value of bruise volume at 1% of statistical level. By increasing temperature، the bruised volume was decreased، whereas it increased by increase of energy level in both varieties. While، the Golden Delicious had more strength than Red Delicious. Also، the maximum rate of bruised volume was related to Red Delicious in contacting to galvanized iron and the minimum rate was related to Golden Delicious in contacting to corrugated carton. Prediction of bruised volume was provided using artificial neural network based on four factors of: temperature، impact energy، padding surface and variety. Multilayer perceptron of neural networks were used for prediction of bruised volume. In comparison with other topologies، algorithm Levenberg-Marquardt had better performance with structure 1-26-4 and logsigmoid transfer function in hidden layer. Based on this algorithm، the mean of prediction accuracy in training، evaluation and testing process was equal to 92. 48، 88. 94 and 87. 72 percent، respectively. In addition، the correlation coefficient (R) was calculated equal to 0. 975 for linear regression between predicted model and experimental data.

Sorption of flavor compounds to inner layer of polymer packages and subsequent diffusion results in loss of food flavor and consequently decreases shelf life and consumer acceptance of food stuffs which in turn causes major economic losses, so it is of utmost importance to research on diffusivity of these compounds in polymers to minimize negative effects of this phenomenon. In current research work, artificial neural networks have been applied to model polymer diffusivity to aroma molecules. The model considers all the factors that play a major role in diffusion including environmental factors, aroma molecule structure and polymer structure and is able to provide diffusion coefficient.This model could be used to predict diffusion coefficient of aroma compounds in high density polyethylene packages. Soft drinks extracts, essential oils, fruit concentrates and many other concentrated products are distributed in high density polyethylene packages. The model is able to calculate shelf life and optimum storage conditions of these products. Meanwhile it eliminates high costs involved in measuring diffusion coefficients of aroma compounds. This model at the range of aroma compounds used in this study functions excellent which is the characteristic of modeling using artificial neural networks.

Pumpkin samples were osmotically dehydrated through three different salt concentrations (5 % w/w, 10% w/w and 15% w/w and 50% w/w) sucrose solution at 25°c and 50°c. Osmotic solution - fruit weight ratio was 1:20. Artificial neural network was used for estimating water loss and solid gain values undergoing osmotic dehydration. Perceptron artificial neural network with gradient descent optimization algorithm was utilized by hyperbolic tangent function for outcomes prediction. The results indicated that artificial neural network with one hidden layer depicted the best fitting when the hidden layer had 10 and 18 neurons for solid gain and water loss respectively. Also, it was found that artificial neural network with 6-6 and 22-22 neurons in two hidden layers stated the optimum outcomes to approximate water loss and solid gain, respectively.

Citrus، especially orange، are of great important among agricultural products in the world. In this study thin-layer drying of orange (var. Thompson) was modeled using artificial neural network (ANN). An experimental dryer was used. Thin-layer of orange slices at five air temperatures (40، 50، 60، 70 & 80 ºC)، three air velocities (0. 5، 1 & 2 m/s) and three thicknesses (2، 4 & 6 mm) were artificially dried. Initial M. C. during all experiments was between 5. 4 to 5. 7 (g/g) (d. b.). Mass of samples were recorded and saved every 5 sec. using a digital balance connected to a PC. MLP with momentum and LM were used to train the ANNS. In order to develop ANN''s models، temperatures، air velocity and time are used as input vectors and moisture ration as the output. Results showed a 3-6-1 topology for thickness of 2 mm، 3-7-1 topology for thickness of 4 mm and 3-5-1 topology for thickness of 6 mm، with LM algorithm and TANSIG activation function were able to predict moisture ratio withof 0. 99906، 0. 99919 and 0. 99930 respectively. The corresponding MSE for this topology were 0. 00013، 0. 00012 and 0. 00009 respectively.