M. Hamdani; M. Taki; M. Rahnama; A. Rohani; M. Rahmati-Joneidabad
Abstract
IntroductionControlling greenhouse microclimate not only influences the growth of plants, but is also critical in the spread of diseases inside the greenhouse. The microclimate parameters are inside air, roof, crop and soil temperature, relative humidity, light intensity, and carbon dioxide concentration. ...
Read More
IntroductionControlling greenhouse microclimate not only influences the growth of plants, but is also critical in the spread of diseases inside the greenhouse. The microclimate parameters are inside air, roof, crop and soil temperature, relative humidity, light intensity, and carbon dioxide concentration. Predicting the microclimate conditions inside a greenhouse and enabling the use of automatic control systems are the two main objectives of greenhouse climate model. The microclimate inside a greenhouse can be predicted by conducting experiments or by using simulation. Static and dynamic models and also artificial neural networks (ANNs) are used for this purpose as a function of the metrological conditions and the parameters of the greenhouse components. Usually thermal simulation has a lot of problems to predict the inside climate of greenhouse and the error of simulation is higher in literature. So the main objective of this paper is comparison between two types of artificial neural networks (MLP and RBF) for prediction 4 inside variables in an even-span glass greenhouse and help the development of simulation science in estimating the inside variables of intelligent greenhouses.Materials and MethodsIn this research, different sensors were used for collecting the temperature, solar, humidity and wind data. These sensors were used in different positions inside the greenhouse. After collecting the data, two types of ANNs were used with LM and Br training algorithms for prediction the inside variables in an even-span glass greenhouse in Mollasani, Ahvaz. MLP is a feed-forward layered network with one input layer, one output layer, and some hidden layers. Every node computes a weighted sum of its inputs and passes the sum through a soft nonlinearity. The soft nonlinearity or activity function of neurons should be non-decreasing and differentiable. One type of ANN is the radial basis function (RBF) neural network which uses radial basis functions as activation functions. An RBF has a single hidden layer. Each node of the hidden layer has a parameter vector called center. This center is used to compare with the network input vector to produce a radially symmetrical response. Responses of the hidden layer are scaled by the connection weights of the output layer and then combined to produce the network output. There are many types of cross-validation, such as repeated random sub-sampling validation, K-fold cross-validation, K×2 cross-validation, leave-one-out cross-validation and so on. In this study, we pick up K-fold cross- validation for selecting parameters of model. The K-fold cross-validation is a technique of dividing the original sample randomly into K sub-samples. Different performance criteria have been used in literature to assess model’s predictive ability. The mean absolute percentage error (MAPE), root mean square error (RMSE) and coefficient of determination (R2) are selected to evaluate the forecast accuracy of the models in this study.Results and Discussion The results of neural networks optimization models with different networks, dependent on the initial random values of the synaptic weights. So, the results in general will not be the same in two different trials even if the same training data have been used. So in this research K-fold cross validation was used and different data samples were made for train and test of ANN models. The results showed that trainlm for both of MLP and RBF models has the lower error than trainbr. Also MLP and RBF were trained with 40 and 80% of total data and results indicated that RBF has the lowest sensitivity to the size data. Comparison between RBF and MLP model showed that, RBF has the lowest error for prediction all the inside variables in greenhouse (Ta, Tp, Tri, Rha). In this paper, we tried to show the fact that innovative methods are simple and more accurate than physical heat and mass transfer method to predict the environment changes. Furthermore, this method can use to predict other changes in greenhouse such as final yield, evapotranspiration, humidity, cracking on the fruit, CO2 emission and so on. So the future research will focus on the other soft computing models such as ANFIS, GPR, Time Series and … to select the best one for modeling and finally online control of greenhouse in all climate and different environment.ConclusionThis research presents a comparison between two models of Artificial Neural Network (RBF-MLP) to predict 4 inside variables (Ta, Tp, Tri, Rha) in an even-span glass greenhouse. Comparison of the models indicated that RBF has lower error. The range of RMSE and MAPE factors for RBF model to predict all inside variables were between 0.25-0.55 and 0.60-1.10, respectively. Besides the results showed that RBF model can estimate all the inside variables with small size of data for training. Such forecasts can be used by farmers as an appropriate advanced notice for changes in temperatures. Thus, they can apply preventative measures to avoid damage caused by extreme temperatures. More specifically, predicting a greenhouse temperature can not only provide a basis for greenhouse environmental management decisions that can reduce the planting risks, but also could be as a basic research for the feedback-feed-forward type of climate control strategy.
F. Jannatdost; P. Ahmadi Moghaddam; F. Sharifian
Abstract
Introduction Fruits and vegetables play an important role in food supply and public health. This group of agricultural products due to high humidity are perishable and most of them (5 to 50 percent) waste during post-harvest operation. Decreasing and minimizing such waste as "hidden harvest" could be ...
Read More
Introduction Fruits and vegetables play an important role in food supply and public health. This group of agricultural products due to high humidity are perishable and most of them (5 to 50 percent) waste during post-harvest operation. Decreasing and minimizing such waste as "hidden harvest" could be an effective way to save food and increase profitability. Despite the surplus of the fruit production in the country, our position in terms of exportation is not commensurate with production, so measurements and grading on the basis of qualitative parameters such as firmness, taste, color, and shape can influence the marketing and export of fruit. In this research, application of an acoustic test is considered to achieve an effective and economic technology in the field to determine the stiffness of kiwifruit in post-harvest step. The aim of this study is to investigate the stiffness index of kiwifruit and provide a classification algorithm in the post-harvest step by using the non-destructive method of processing impact acoustic signals. Materials and Method In this research, an acoustic-based intelligent system was developed and the possibility of using the acoustic response to classify kiwifruit into soft, semi-soft and stiff categories was studied. 150 samples of Hayward variety of Kiwifruit was used during the 18 days shelf life in controlled conditions of temperature and humidity. Analyses were done in 9 sets per two days. In each analysis, an acoustic test was done by 48 samples in both free fall condition and fall from a conveyor belt. The feature extraction of acoustic signals in both the time domain and frequency domain has done, then the classification of samples was done by using the Artificial Neural Network. After getting the impact signals of stiff, semi-soft and soft samples, stiffness of kiwifruits identification has done by using acoustic features. The stiffness of kiwifruit samples in this study was measured to be 15.9±4.9 (N) by using the Magnes- Taylor test. Finally, samples were classified into stiff, semi-soft and soft by comparison of maximum force and flux of signals amplitude. Results and Discussion The results showed that the features of CF and maximum amplitude in the time domain have high accuracy in kiwifruit classification. The frequency resonances as environmental noises or impact position are out of control in the time domain which causes a decrease in accuracy. So, the ANN by features of time domain has not the acceptable capability to identify the semi-soft samples. The identification of semi-soft samples is not easy because of having same properties of stiff and soft samples. Extracted features of frequency domain have the most capability of correct detection. The optimal network has five neurons in the hidden layer and 0.014782 of mean square error. The accuracy of correct detection of the optimal network was 93.3, 91.3 and 78.3 percent for stiff, semi-soft and soft samples, respectively. Because of using more features in the frequency domain, the classification of all categories was acceptable and identification of semi-soft samples was as good as stiff and soft samples. The results of combined features of time and frequency domain showed that the artificial neural network has less efficiency in comparison with the other two attitudes. The accuracy of identification and classification was decreased by adding the extracted features of the time domain. So achieving the most accuracy in classification is accomplishable just by using the features of the frequency domain. By comparing the results of both free fall and online tests, it is claimed that this research can be industrialized. Conclusion Comparison of all results shows that there was no significant difference in the capability of ANN for identification and classification of the sample in three categories. After all, we can use this method in online sorting of kiwifruits by controlling the vector and position of impaction.
S. I. Saedi; R. Alimardani; H. Mousazadeh
Abstract
Introduction Global solar radiation is the sum of direct, diffuse, and reflected solar radiation. Weather forecasts, agricultural practices, and solar equipment development are three major fields that need proper information about solar radiation. Furthermore, sun in regarded as a huge source of renewable ...
Read More
Introduction Global solar radiation is the sum of direct, diffuse, and reflected solar radiation. Weather forecasts, agricultural practices, and solar equipment development are three major fields that need proper information about solar radiation. Furthermore, sun in regarded as a huge source of renewable and clean energy which can be used in numerous applications to get rid of environmental impacts of non-renewable fossil fuels. Therefore, easy and fast estimation of daily global solar radiation would play an effective role is these affairs. Materials and Methods This study aimed at predicting the daily global solar radiation by means of artificial neural network (ANN) method, based on easy-to-gain weather data i.e. daily mean, minimum and maximum temperatures. Having a variety of climates with long-term valid weather data, Washington State, located at the northwestern part of USA was chosen for this purpose. It has a total number of 19 weather stations to cover all the State climates. First, a station with the largest number of valid historical weather data (Lind) was chosen to develop, validate, and test different ANN models. Three training algorithms i.e. Levenberg – Marquardt (LM), Scaled Conjugate Gradient (SCG), and Bayesian regularization (BR) were tested in one and two hidden layer networks each with up to 20 neurons to derive six best architectures. R, RMSE, MAPE, and scatter plots were considered to evaluate each network in all steps. In order to investigate the generalizability of the best six models, they were tested in other Washington State weather stations. The most accurate and general models was evaluated in an Iran sample weather station which was chosen to be Mashhad. Results and Discussion The variation of MSE for the three training functions in one hidden layer models for Lind station indicated that SCG converged weights and biases in shorter time than LM, and LM did that faster than BR. It means that SCG provided the fastest performance. However, the story for accuracies was different i.e. the BR, LM, and SCG algorithms provided the most accurate performances, respectively, both among one or two hidden layers. According to the evaluation criteria, six most accurate derived models out of 1260 tested ones for Lind station was 3-14-1 and 3-11-19-1 with LM, 3-20-1 and 3-20-19-1 with BR, and 3-9-1 and 3-20-17-1 with SCG training algorithm, and 3-20-19-1 topology with BR showed the best performance out of all architectures. Results of the evaluation of the six accurate models in the remaining 18 stations of Washington State proved that regardless of the climate, in each weather station, BR with its inherent automatic regularization, provided the most accurate models (0.87 67.41 %), and then SCG (0.90>R>0.83, 3.91>RMSEMAPE > 77.28 %). Therefore, the Bayesian neural networks, which showed the best performance among all Washington State weather stations, were evaluated for Mashhad station, as an Iran sample climate. The results proved the ability of the said networks for this climate (R=0.82, RMSE=3.92 MJm-2, MAPE=79.92%). Conclusion The results indicated that the Bayesian neural networks are capable of predicting global solar radiation with minimum inputs in different climates. This was concluded both in Washington State weather stations, which has a variety of climates, and also in Mashhad as an Iran sample weather station. These models would eliminate the need for complex climate-dependent mathematical relations or other models which are mostly dependent on many inputs. So, this algorithm would be a good means first in weather forecast practices, also in the design and development of solar assisted equipment, as well as in managerial practices in agriculture when monitoring crop solar-dependent processes like photosynthesis and evapotranspiration.
S. F. Mousavi; M. H. Abbaspour-Fard; M. H. Aghkhani; E. Ebrahimi; A. Soheili Mehdizadeh
Abstract
Introduction
The diagnosis of agricultural machinery faults must be performed at an opportune time, in order to fulfill the agricultural operations in a timely manner and to optimize the accuracy and the integrity of a system, proper monitoring and fault diagnosis of the rotating parts is required. ...
Read More
Introduction
The diagnosis of agricultural machinery faults must be performed at an opportune time, in order to fulfill the agricultural operations in a timely manner and to optimize the accuracy and the integrity of a system, proper monitoring and fault diagnosis of the rotating parts is required. With development of fault diagnosis methods of rotating equipment, especially bearing failure, the security, performance and availability of machines has been increasing. In general, fault detection is conducted through a specific procedure which starts with data acquisition and continues with features extraction, and subsequently failure of the machine would be detected. Several practical methods have been introduced for fault detection in rotating parts of machineries. The review of the literature shows that both Artificial Neural Networks (ANN) and Support Vector Machines (SVM) have been used for this purpose. However, the results show that SVM is more effective than Artificial Neural Networks in fault detection of such machineries. In some smart detection systems, incorporating an optimized method such as Genetic Algorithm in the Neural Network model, could improve the fault detection procedure. Consequently, the fault detection performance of neural networks may also be improved by combining with the Genetic Algorithm and hence will be comparable with the performance of the Support Vector Machine. In this study, the so called Genetic Algorithm (GA) method was used to optimize the structure of the Artificial Neural Networks (ANN) for fault detection of the clutch retainer mechanism of Massey Ferguson 285 tractor.
Materials and Methods
The test rig consists of some electro mechanical parts including the clutch retainer mechanism of Massey Ferguson 285 tractor, a supporting shaft, a single-phase electric motor, a loading mechanism to model the load of the tractor clutch and the corresponding power train gears. The data acquisition section consists of a data analyzer (PCA-40), a personal computer, a piezoelectric accelerometer (VMI-102, DT-2234B), a tachometer and two rubber vibration absorbing elements are located between the rig’s components and the plate holder. An evaluation function was employed in order to achieve the optimal structure of neural network models by selecting the number of layers, number of cells in the layers, transfer function, training function, learning functions, performance function, and number of epochs, in such a way that the MSE of the calculated output error was minimal. The data were collected by means of the accelerometer sensor attached on the clutch mechanism, with three different working conditions (normal condition, with worn bearing, and with worn shaft), and three rotational speeds including: 1000 rpm, 1500 rpm and 2000 rpm. The Wavelet Packet Transform (WPT) was applied on the data-set for features vector extraction and the principle component analyses (PCA) was applied for dimension reduction of the features vector. The signal processing and the features extraction are the most important characteristics of the monitoring methodology, by which the working condition of the machine can be determined. These characteristics may be acquired by transforming the signals from the time domain to the frequency domain and MATLAB software is used for this purpose. This software receives the vibration data (time series of output voltage) which are in Excel files format. To remove the noise a suitable filtering procedure was used and finally the statistical parameters of time - frequency were calculated.
Results and Discussion
To verify the accuracy of the Genetic Algorithm model, the required data were collected from the training and testing steps of the Neural Network. For this purpose, the statistical parameters such as mean squared error (MSE), mean absolute error (MAE) and correlation coefficient (r) were used. The optimal parameters of the neural network obtained for the family of Db4. A trial and error procedure was used to minimize the mean square error of the network output and the desired amount of training step. During the training step, four neural networks including Db4, Db30, Db35 and Db40 achieved a gradient descent weight in the learning bias and four neural networks including Db9, Db15, Db20 and Db25 achieved a gradient descent with momentum weight in the learning bias. The two of the achieved neural networks including Db4, Db20 have circular logarithm function and the remaining networks have annular hyperbolic tangent transfer function. The most appropriate networks configuration was acquired when the network exhibited the minimal error with the training and testing data sets. The results show that the highest accuracy of the GA-ANN Artificial neural networks for all rotational speeds (1000, 1500 and 2000 rpm), and working conditions (intact gear and shaft, damaged bearing and worn shaft) observed for the network family of Db4. The highest error observed for the family of Db20 with MSE of 0.011.
Conclusions
Artificial neural networks can somewhat think and make decisions similar to an expert person. In this project in order to predict the occurrence of a failure of the clutch mechanism of MF 285 tractor, the experimental data were obtained using some sensors, and the data were transferred to a computer by means of a data analytical. By training of the neural networks, the errors were identified separately. The output data from the combined Neural Network and Genetic Algorithm shows that the performance of the prediction model is enhanced. Based on the experiments and calculations, the best data set belongs to the family of Db4 network with the least MSE equal to 4.09E-07 and r equal to 0.99999, indicating that the model could precisely detect the faulty bearings or shafts.
Design and Construction
S. Khalili; B. Mohammadi Alasti; M. Abbasgholipour
Abstract
Introduction: Grading agricultural products always has a particular important position for submission to domestic and overseas markets. The grading causes more profitable product ranges and customer satisfaction. Grading treatment is carried out based on various parameters such as color, ripeness level, ...
Read More
Introduction: Grading agricultural products always has a particular important position for submission to domestic and overseas markets. The grading causes more profitable product ranges and customer satisfaction. Grading treatment is carried out based on various parameters such as color, ripeness level, dimensions and weight. Product weight is one of the most effective parameters in grading operation. Egg weight is directly related to the smallness and coarseness of eggs. In egg grading, the largeness value is very important in marketing. This research aimed to design, fabricate and evaluate the egg weighing system based on its dielectric properties.Materials and Methods: To perform this research, the stages of work are divided into several sections including, design and construction of the hardware section, writing code for the software section to collect data, conducting nondestructive tests and data collection, analysis of obtained data using artificial intelligence, and giving the results of analysis for device calibration of the system as the software code. The large eggs as dielectric substances cause more increase in the capacity of the capacitive sensor. Furthermore, by derivation of a relation between capacity of capacitive sensor and egg weight, one can predict the weight of the sample. A prototype unit of weighing system was designed and fabricated. The designed unit was composed of a chassis, a voltage source, a sinusoidal signal generator, a voltage measurement unit, an AVR micro controller, a COM port, a capacitive sensor, and an LCD and a keyboard. Neural network technique was used for egg weight prediction. The designed net receives 16 voltage values at different frequencies as inputs and its output is the egg weight. In order to calibrate and evaluate the weighing unit, 150 fresh egg samples were provided on egg laying day from a local poultry farm. Experiments were divided into three groups. The experiments were carried out on egg-laying day, and the second and fourth day after laying.Results and Discussion: In this study, two networks were built and evaluated. In the first series, two-layer networks and in the second series, three-layer networks were developed. In the two-layer neural networks, the number of neurons in the hidden layer was changed from 2 to 10.According to the given results for two-layer networks, two layer networks with 10 neurons offer the best results (the highest R-value and minimum RMSE) and it can be chosen as the most effective two-layer network. Three-layer neural networks have been composed of two hidden layers. The number of neurons in the first hidden layer was 10 and in the second layer it was changed from 1 to 20. Between three-layer networks, the network with 7 neurons with the highest R-value and the lowest error is the most appropriate network. It is even more efficient than the two-layer network with 10 neurons. So, the most appropriate structure is 1-7-10-16 and it has been selected for calibration of the weighing device. To evaluate and assess the accuracy of the weighing machine, weights of 24 samples of fresh eggs were predicted and compared with the actual values obtained using a digital scale with the accuracy of 0.01 gr. The paired t-test has been used to compare the measured and predicted values and the Bland-Altman method has been used for charting the accordance between the measured and predicted values. Based on the findings, the difference between the measured and predicted values was observed up to 5.4 gr that is related to a very large sample. The mean absolute error is equal to 2.21 gr and the mean absolute percentage error is equal to 3.75 %. According to the findings, 95% of the actual and approximate matching range to compare the two weighing methods is between -5.3 gr and 3.36 gr. Thus, the dielectric technique may underestimate the egg weight up to 5.3 gr or it may overestimate it up to 3.36 gr more than the actual prediction.Conclusions: The best results were obtained with a 3 layers net having 10 and 7 neurons, respectively in the first and the second hidden layers with the highest R-value, 0.983 and the lowest error, 0.502. Therefore, this net was applied for egg weight prediction. To evaluate the device, the weights of 24 fresh eggs were estimated using the device and were compared with actual values and the maximum error was observed to be equal to 5.4 gr.
A. Rohani; S. I. Saedi; H. Gerailue; M. H. Aghkhani
Abstract
Introduction: Fast and accurate determination of geometrical properties of agricultural products has many applications in agricultural operations like planting, cultivating, harvesting and post-harvesting. Calculations related to storing, shipping and storage-coating materials as well as peeling time ...
Read More
Introduction: Fast and accurate determination of geometrical properties of agricultural products has many applications in agricultural operations like planting, cultivating, harvesting and post-harvesting. Calculations related to storing, shipping and storage-coating materials as well as peeling time and surface-microbial concentrations are some applications of estimating product volume and surface area. Sphericity is also a parameter by which the shape differences between fruits, vegetables, grains and seeds can be quantified. This parameter is important in grading systems and inspecting rolling capability of agricultural products. Bayram presented a new dimensional method and equation to calculate the sphericity of certain shapesand some granular food materials (Bayram, 2005). Kumar and Mathew proposed atheoretically soundmethod for estimating the surface area of ellipsoidal food materials (Kumar and Mathew, 2003). Clayton et al. used non-linear regression models for calculation of apple surface area using the fruit mass or volume (Clayton et al., 1995). Humeida and Hobani predicted surface area and volume of pomegranates based on the weight and geometrical diametermean (Humeida and Hobani, 1993). Wang and Nguang designeda low cost sensor system to automatically compute the volume and surface area of axi-symmetricagricultural products such as eggs, lemons, limes and tamarillos (Wang and Nguang, 2007). The main objective of this study was to investigate the potential of Artificial Neural Network (ANN) technique as an alternative method to predict the volume, surface area and sphericity of pomegranates.
Materials and methods: The water displacement method (WDM) was used for measuring the actual volume of pomegranates. Also, the sphericity and surface area are computed by using analytical methods. In this study, the neural MLP models were designed based upon the three nominal diameters of pomegranatesas variable inputs, while the output model consisted of each of the three parameters including the volume, sphericity and surface area. Priorto any ANN training process, the data normalized over the range of [0, 1]. Fig. 1 shows a MLP with one hidden layer. In this study, back-propagation with declininglearning-rate factor (BDLRF) training algorithm was employed. The mean absolute percentage error (MAPE) and the coefficient of determinationof the linear regression line between the predicted values fromthe MLP model and the actual output were used to evaluate the performance of the model.
Results and Discussion: The number of neurons in the hidden layerand also theoptimal values for the learning parameters η and αwere selected bytrial and error method. The bestresult was achieved with five neurons in the hidden layer. The results showed thatthe optimum modelof performance was obtained at constant momentum termequal to 0.8 and learning rate equal to 0.9. In this study, 300 epochs were selected as the starting points of the BDLRF. Some statistical characteristics of the actual values of volume were estimated by WDM, surface area was computed by equation (3) and sphericity of pomegranates was computed by equation (1) and the predicted values of them using the neural network method were shown in Table 1. The obtained results verified that the differences between theactual values and the estimated ones can be ignored. But, the predicted values of the volume using the MLP model in comparison with equation (2) are much closer to the actual values. Statistical comparisons of desired and predicted data and the corresponding p values are given in Table 2. The results showed that P-value was greater than 0.08 in all cases. Therefore, there was no significant difference between the statistical parameters. However, the P-value for equation 2 is much less than that of the MLP model. The results shown in Figures 2, 3 and 4 show that the coefficients of determination between actual and predicted data were greater than 0.9. Considering all the results in our study, the MLP model is more accurate than the WDM and analytical methods.
Conclusions: In this paper, we first measure the actual volume of the pomegranate using WDM and equation (2). Also, assuming an elliptical fruit, the sphericity and surface area are computed analytically based on the three nominal diameters of a pomegranate. Finally, the results of achievements of the MLP designed revealed that the MLP model could be successfully applied to the prediction of thesphericity and surface area. Therefore, the MLP model can be a viable alternative to the analytical methods. However, this is possible only if there is a precise way to compute the three nominal diameters of pomegranates. In addition, according to the MAPE, the accuracy of the MLP model in prediction of volume of pomegranates was twicethe analytical method.
Image Processing
M. Jafarlou; R. Farrokhi Teimourlou
Abstract
Physical properties of agricultural products such as volume are the most important parameters influencing grading and packaging systems. They should be measured accurately as they are considered for any good system design. Image processing and neural network techniques are both non-destructive and useful ...
Read More
Physical properties of agricultural products such as volume are the most important parameters influencing grading and packaging systems. They should be measured accurately as they are considered for any good system design. Image processing and neural network techniques are both non-destructive and useful methods which are recently used for such purpose. In this study, the images of apples were captured from a constant distance and then were processed in MATLAB software and the edges of apple images were extracted. The interior area of apple image was divided into some thin trapezoidal elements perpendicular to longitudinal axis. Total volume of apple was estimated by the summation of incremental volumes of these elements revolved around the apple’s longitudinal axis. The picture of half cut apple was also captured in order to obtain the apple shape’s indentation volume, which was subtracted from the previously estimated total volume of apple. The real volume of apples was measured using water displacement method and the relation between the real volume and estimated volume was obtained. The t-test and Bland-Altman indicated that the difference between the real volume and the estimated volume was not significantly different (p>0.05) i.e. the mean difference was 1.52 cm3 and the accuracy of measurement was 92%. Utilizing neural network with input variables of dimension and mass has increased the accuracy up to 97% and the difference between the mean of volumes decreased to 0.7 cm3.
A. Rohani; H. Makarian
Abstract
With the rise of new powerful statistical techniques and neural networks models, the development of predictive species distribution models has rapidly increased in ecology. In this research, a learning vector quantization (LVQ) and multi layer perceptron (MLP) neural network models have been employed ...
Read More
With the rise of new powerful statistical techniques and neural networks models, the development of predictive species distribution models has rapidly increased in ecology. In this research, a learning vector quantization (LVQ) and multi layer perceptron (MLP) neural network models have been employed to predict, classify and map the spatial distribution of A. repens L. density. This method was evaluated based on data of weed density counted at 550 points of a fallow field located in Faculty of Agriculture, Shahrood University of Technology, Semnan, Iran, in 2010. Some statistical tests, such as comparisions of the means, variance, statistical distribution as well as coefficient of determination in linear regression were used between the observed point sample data and the estimated weed seedling density surfaces by two neural networks to evaluate the performance of the pattern recognition method. Results showed that in the training and test phases non significant different was observed between average, variance, statistical distribution in the observed and the estimated weed density by using LVQ neural network. While this comparisions was significant except statistical distribution by using MLP neural network. In addition, results indicated that trained LVQ neural network has a high capability in predicting weed density with recognition erorr less than 0.64 percent at unsampled points. While, MLP neural network recognition erorr was less than 14.6 percent at unsampled points. The maps showed that, patchy weed distribution offers large potential for using site-specific weed control on this field.
