جستجوی مقالات مرتبط با کلیدواژه "شبکه عصبی مصنوعی" در نشریات گروه "گیاهپزشکی"

Nowadays, many researchers have paid attention to explaining the patterns of insect dispersion using interpolation and density estimation methods in order to investigate the possibility of proper integrated management with the location of pests. This study was conducted to predict and map the spatial distribution of the Coccinella septempunctata using multilayer perceptron neural networks (MLPs) combined with the imperialist competitive algorithm and shuffled frog leaping algorithm at the field level. Data have been obtained through100 samples taking from the surface of a hay field of zarghan area in 2019. To evaluate the neural networks used and compare their performance, statistical parameters such as statistical distribution, mean comparison and coefficient of explanation between the spatially predicted values by the neural network and their actual values were used to predict the distribution of this species. The results showed that in the training and experimental phases, there was no significant difference between the values of statistical distribution and the mean of real and predicted spatial data sets of this species combined by neural network with shuffled frog leaping algorithm. Shuffled frog leaping algorithm was more accurate in detecting the distribution. Our map showed that pest distribution was patchy

In this research, Quantitative Structure–Activity Relationship (QSAR) study has been used for prediction of toxicity values of carboxylic acid derivatives. Firstly, the toxicity (LD50) values of data set of studied compounds were taken from the scientific web book and the their structures were drawn with the Gauss view 05 program and optimized at Hartree–Fock level of theory and 3-21G basis set by Gaussian 09 software. Then the dragon software was used for the calculation of molecular descriptors. The unsuitable descriptors were deleted with the aid of the genetic algorithm (GA) and backward techniques, and the best descriptors were used for multiple linear regression (MLR) and artificial neural network (ANN) models. The prediction accuracy of the final model was discussed using the statistical parameters. Leave-one-out cross-validation and external test set of the predictive models demonstrated a high-quality correlation between the observed and predicted toxicity values of all, training, test and validation sets in GA-ANN method. The model by ANN algorithm due to the lower error and higher regression coefficients was clearly superior to those models by MLR algorithm. The proposed model may be useful for predicting log LD50 of new compounds of similar class.

Regarding the importance of greenhouse productions and high activity of pests including Trialeurodes vaporariorum under greenhouse conditions, the management of this pest requires ecological studies with a new approach. Therefore, due to the influence of the host plant characteristics on biological performance of greenhouse whitefly, the current research was performed to predict and estimate the values of its population growth parameters including net reproduction rate (R0) and intrinsic rate of population increase (rm). Estimation was based on some morphological features of the host plants using a MLP artificial neural network. The network was optimized with a genetic algorithm. The R0 and rm values of T. vaporariurum were calculated on two host plants, Cucumis sativus L. and Cucumis metuliferus May. Moreover, density and length of the leaf trichomes, density and area of leaf stomata cell of the lower leaf surface and the amount of leaf chlorophyll of each host plant was measured. The MLP neural network with optimal algorithm was designed. In order to evaluate the MLP neural network the T-test, F-test and Kolmogorov-Smirnov test were used to compare mean, variance, and statistical distribution, respectively. The obtained coefficient of determination (R2 = 0.9621) and probability level (P > 0.773) of statistical tests indicated high accuracy and capability and high generalizability of the MLP neural network for estimating R0 and rm of greenhouse whitefly.

Nowadays, with the advent of powerful statistical techniques and neural networks, predictive models of distribution have been rapidly developed in ecology. This study was carried out to model distribution of aphid, Rhopalosiphum padi,using MLP neural networks combined with Particle Swarm Optimization in wheat fields of Badjgah area, Fars province. Population data of the pest was obtained by sampling at 100 locations across wheat fields during 2013. For evaluation the capability of neural networks used in dispersal prediction, statistical comparison of parameters such as mean, variance, statistical distribution of spatial predicted values by neural network and their actual values, were conducted. Results showed that there were not significant differences between variance, mean and statistical distribution of actual and predicted values in training and test phases of neural network combined Particle Swarm Optimization algorithm. Our map showed a patchy pest distribution offers large potential for using site-specific pest control on this field.

Due to the high speed and accuracy of intelligent pest detection in warehouse products, in this study, the detection of chickpea four-point beetle pest was simulated by image processing technique using artificial neural networks. To prepare the images, a glass box was prepared and the chickpea seeds were placed in the center of the box. The light was then illuminated from all six sides and photographed with a digital camera from all sides. The image properties were then extracted by Wavelet Gabor using MATLAB software and applied to the ANN as training data. To train the network, 69 images of chickpeas damaged and 59 healthy chickpeas were used. Then, to evaluate the network, a set of data that did not play a role in network training as test data was applied to the network and its results were evaluated. In this study, Perceptron and Elman neural networks were used which had better results than Elman network. The proposed method was able to detect the high rate of damaged with 6.17% non-detection error and 4.86% error-detection error. After image processing by the neural network and detection of damage points, the amount of crop damage was also calculated. For this purpose, the level of detected damage was calculated and divided by the area of total area of chickpea seed and percentage of damage. After identifying the injury sites, the damage was estimated 2.3% in the studied images.

In this research, a learning vector quantization neural network (LVQ) model was developed to predict the spatial distribution of Sitona humeralis in Marvdasht. This method was evaluated on data of pest density from alfalfa field. Pest density assessments were performed following a 10 m × 10 m grid pattern on the field and a total of 100 sampling units on field. Some statistical tests, such as means comparison, variance and statistical distribution were performed between the observed points samples data and the estimated pest values to evaluate the performance of prediction of pest distribution. The Results showed that in training and test phase, there were no significant differences, with the confidence level of 95%, between the statistical parameters such as average, variance, statistical distribution and also coefficient of determination in the observed and the estimated pest density. The results suggest that learning vector quantization (LVQ4) neural network can learn pest density model precisely. In addition the results also indicated that trained LVQ4 neural network had a high capability (88%) in predicting pest density for non-sampled points. The technique showed that the LVQNN could predict and map the spatial distribution of Sitona humeralis. The map showed that the pest has aggregation distribution so there is possibility potential for using site-specific pest control on this field.

The organophosphate pesticides are considered as dangerous chemical pesticides for human health. The possibility of absorbing these toxins through the skin is accessible to many researchers who have been studying the toxicity of organophosphate compounds. Experimental methods are time-consuming and high cost, and they come with a lot of dangers. Quantitative structure activity/property relationship studies provide the capability to access data, information and physico-chemical properties of chemical compounds, using the methods and modeling.  In this study, the multiple regression linear method and the artificial neural network with multi-layer perceptron (MLP) model were used to investigate the quantitative relationship of LD50 </sub>(mgkg-1</sup>) toxicity index with some molecular descriptors of some organophosphate compounds.  Investigation of correlation coefficients and root mean square errors values of final models in this study showed that ANN method using the MLP model was higher than the MLR method for prediction of LD50</sub>(mgkg-1</sup>) of organophosphates compounds.

In recent decades, computational methods with regard to accurate validation parameters for the determination of the physical- chemistry properties of compounds have been considered by many researchers and have been used as an economic and environmental alternative to saving time and eliminating high costs.  In this study, the relationship between the logarithmic values of LD50</sub>,) log (LD50</sub>)(molkg-1</sup>) and molecular descriptors has been investigated for 60 types of aniline derivatives(including some herbicides compounds). At first, the structure of the compounds were drawn by Gauss view05 software and optimized using Gaussian 09 software with B3LYP/6-311++G** method, and then were extracted molecular descriptors. Then inappropriate descriptors were eliminated by genetic algorithm method and the best ones were used for multiple linear regression (MLR) and artificial neural networks (ANN) models. The results showed that the ANN method with the lowest error and the highest coefficient of determination was higher than the MLR method to predicting the log(LD50</sub>)(molkg-1</sup>) of studied aniline derivatives.

The Sunn pest, Eurygaster integriceps Put. is the most important pest of wheat and barley in Iran. Many studies have shown that various biotic and abiotic environmental factors affect the population of this pest. In this study, the relationship between population density of the Sunn pest adult insect with different environmental factors including sampling date, average daily temperature, average relative humidity, wind speed, wind direction, height from sea level and degree-day was investigated. Field data were collected from two wheat farm of one-hectare in the city of Chadegan, Isfahan province. The used network type was multilayer perceptron with back propagation algorithm and the learning algorithm was Levenberg-Markvart. After sensitivity analysis due to the ease of the model and extraction of effectiveness of factors including four factors of sampling date, temperature, humidity and wind speed were selected. The results showed that a neural network with two hidden layer, 7 neuron in the first hidden layer and three neuron in the second hidden layer, as a sigmoid activation function, and a data percentage of 60, 30, 10 for training, testing and validation for prediction of population fluctuation of the Sunn pest adult insect is used (R2= 0.94).

Today, with the advance statistical techniques and neural networks, predictive models of distribution were rapidly developed in ecology. Due to the difficulty of sampling, there are usually not enough samples in such studies. Therefore, in order to predict and mapping the distribution of Coccinella
septempunctata used the combination of the Kriging method with multilevel perceptron neural networks (MLP) combined with genetic algorithm at the farm level. Population data of pest was obtained in 2014 by sampling in 221 fixed points in the alfalfa farm of Bajgah. The data was interpolated by ordinary Kriging method with spherical variogram, which had the best performance, and introduced as a neural network input. To evaluate the ability combined geostatistics with optimized artificial neural network by genetic to predict the distribution used statistical comparison parameters such as mean, variance, statistical distribution and between predicted values and actual values. Results indicating that there was non-significant difference between statistical parameters such as average, variance, statistical distribution and also coefficient of determination in the observed and the estimated Coccinella septempunctata density. Our map showed that pest distribution was patchy.

This study aimed to predict population fluctuation of sunn pest in the field using artificial neural network and multiple linear regression was performed. The data on population fluctuation of Sunn pest in years 2015 and 2016 on a farm with an area of one hectare in the city Chadegan was obtained. In this model of the variables sampling date, the average temperature, average relative humidity, wind speed, wind direction, rainfall as the input variables and population changes mother Sunn pest was used as the outcome variable. The network was used of type Multilayer Perceptron with back propagation algorithm and was learning method Levenberg Markvart. Results showed between these two models, artificial neural network with coefficient of determination 0.96 better than regression with coefficient of determination 0.40 population density of mother Sunn pest was predicted. After sensitivity analysis model for easier and factors more effective extraction, four factors: the number of days of the year, temperature, humidity and wind speed were selected. Neural network model was trained again using the four factor model and a model with 11 hidden layer gave the best result. The coefficient of determination testing stepe was 0.97 that was showed high accuracy relative to the multiple linear regression model with the coefficient of determination 0.43.

Plant disease can cause quality and quantity reduction of agriculture crops. In some countries farmers spend considerable time to consult with plant pathologists, whereas time is an important factor to control disease, so it seems to offer a fast, cheap and accurate method to detect plant diseases. Since the fungal disease named ‘Powdery Mildew’ and ‘Anthracnose’ cause the greatest amount of damage in cucumber produced in greenhouses, thus in this research the mentioned two fungal disease detection and classification were studied using image processing and neural networks technique. Image processing steps includes four main steps: 1) Image acquisition 2) preprocessing 3) extraction of the best color parameters of HSV and L*a*b* color spaces in order to classify and extract defected area of the leaf and 4) extraction of textural properties of defected area of cucumber leaf using co-occurrence matrix. Since, two factors of accuracy and time are important in detection and classification of plant disease, thus artificial neural networks (ANN) with back propagation algorithm (BP) and Levenberg-Marquardt (LM) training function was selected as the best model that was able to successfully detect and classify mentioned plant diseases in 6 seconds with 99.96% accuracy.

In this study, the geostatistical and artificial neural network methods were used to estimate the spatial distribution of Tetranychus urticae in Ramhormoz Cucumber fields. For this purpose, latitude and longitude of 100 points with 10 meters distance of each point were defined as inputs and output of each method was number of these pests on those points. Ordinary kriging, and perceptron with propagation algorithm were evaluated in geostatistical and artificial neural network method, respectively. In neural network a hidden layer and three-layer were considered as input. Results of the aforementioned two methods showed that artificial neural network capability is more than kriging method. So that, the artificial neural network predicts distribution of this pest with 0.891 coefficient of determination and 0.14 residual sums of squares. While in the geostatistical methods coefficient of determination and residual sums of squares were 0.601 and 0.071, respectively. So it can be concluded that the Artificial Neural Network approach with combining latitude and longitude can forecast pest density with sufficient accuracy.

Today, with the Advance statistical techniques and neural networks, predictive models of distribution was rapidly developed in Ecology. Purpose of this study was predict and Mapping distribution of Bemisia tabaci G. using MLP neural networks combined with Particle Swarm Optimization in surface of cucumber field. Population data of pest was obtained in 2017 by sampling in 100 fixed points in a fallow field in Ramhormoz, to evaluate the ability of neural networks combined with Particle Swarm Optimization to predict the distribution used statistical comparison parameters such as mean, variance, statistical distribution and coefficient determination of linear regression among predicted values and actual values. Results showed that in training and test phases of neural network combined Particle Swarm Optimization algorithm, was no significant effect between variance, mean and statistical distribution of actual values and predicted values. Our map showed that patchy pest distribution offers large potential for using site-specific pest control on this field.

In this research, a Learning Vector Quantization (LVQ) neural network model was developed to predict the spatial distribution of Tuta absoluta in tomato fields of the city of Ramhormoz, Iran. Pest density was assessed through 10 m × 10 m grid pattern on the field with a total of 100 sampling units. Some statistical tests, such as means comparison, variance and statistical distribution were performed between the sampling point data and the estimated pest values in order to evaluate the performance of prediction of pest distribution. In training and test phase, there was no significant difference in average, variance, statistical distribution and coefficient of determination at 95% confidence level. The results suggest that LVQ neural network can learn pest density model precisely and trained LVQ neural network high capability (88%) of predicting pest density for non-sampled points. The LVQNN successfully predicted and mapped the spatial distribution of Tuta absoluta whose aggregation distribution implied the possibility of using site-specific pest control in the field.