alexa Irrigation Time Detection System (ITDS): Irrigation Scheduling Using Explicit Neural Network Modeling of Plant and Environmental Variables | Open Access Journals
ISSN: 2168-9768
Irrigation & Drainage Systems Engineering
Make the best use of Scientific Research and information from our 700+ peer reviewed, Open Access Journals that operates with the help of 50,000+ Editorial Board Members and esteemed reviewers and 1000+ Scientific associations in Medical, Clinical, Pharmaceutical, Engineering, Technology and Management Fields.
Meet Inspiring Speakers and Experts at our 3000+ Global Conferenceseries Events with over 600+ Conferences, 1200+ Symposiums and 1200+ Workshops on
Medical, Pharma, Engineering, Science, Technology and Business

Irrigation Time Detection System (ITDS): Irrigation Scheduling Using Explicit Neural Network Modeling of Plant and Environmental Variables

Naeenee AE1*, Boroujeni FZ2, Kimiaei M2, Golabadi M1, Baratia M1, Jahangardb S2 and Mokhtarzadeha A1

1Department of Agriculture, Islamic Azad University, Isfahan (Khorasgan) Branch, Isfahan, Iran

2Department of Computer Engineering, Islamic Azad University, Isfahan (Khorasgan) Branch, Isfahan, Iran

*Corresponding Author:
Naeenee AE
Department of Agriculture, Islamic Azad University
Isfahan (Khorasgan) Branch, Isfahan, Iran
Tel: 0098 9133155791
E-mail: a.eghtedar@gmail.com

Received Date: January 30, 2017; Accepted Date: Febrauary 14, 2017; Published Date: Febrauary 21, 2017

Citation: Naeenee AE, Boroujeni FZ, Kimiaei M, Golabadi M, Baratia M, et al. (2017) Irrigation Time Detection System (ITDS): Irrigation Scheduling Using Explicit Neural Network Modeling of Plant and Environmental Variables. Irrigat Drainage Sys Eng 6: 178. doi: 10.4172/2168-9768.1000178

Copyright: © 2017 Naeenee AE, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Visit for more related articles at Irrigation & Drainage Systems Engineering

Abstract

This study was conducted in four phases to construct models through them the plant water status can be estimated. These relations and models were implemented as a software and embed into a hardware device. The data were collected in the greenhouse and open field conditions during 2010-2015 and modeled by a multilayer feed forward neural network using MATLAB software. In the first phase, 7 variables including vapor pressure deficit, photosynthetically active radiation, wet bulb temperature, leaf temperature (Tl), air temperature (Ta), Ta and Tl difference (Ta-Tl) and relative humidity (RH) were measured for tomato and cucumber. High coefficient of determination (R) values (more than 0.7) was obtained in all fitted models when all variables are incorporated. However, starting from a complete set of variables, the result of stepwise backward elimination of variables showed that removing VPD, PAR, WbT and Tl has a trivial effect on R values. Therefore in the second phase, only 3 variables (Ta, Ta-Tl and RH) were measured to fit the model for 4 greenhouse and 10 open field crops. For each crop, the best model with the highest R value was obtained. These models were used to design an Irrigation Time Detection System (ITDS) in the third phase. The main parts of the system consist of an artificial neural network (ANN) module as the intelligent part of this system implemented on a cell phone coupled with a microcontroller based hardware device. Finally, in the fourth phase, ITDS was tested successfully on tomato and cucumber in the greenhouse.

Keywords

Leaf temperature; Irrigation scheduling; Neural network; Crop water stress

Introduction

Assessment of crop water requirement through soil moisture or crop evapotranspiration is a time-consuming and complicated process and may not give the proper estimation of crop water requirement [1]. In addition, measuring the soil moisture and estimation of the appropriate irrigation time via evapotranspiration are difficult tasks for most of the growers. This could be more crucial in soilless media, where plants should irrigate repeatedly (due to a restricted root system) during a day [2]. Therefore in the past few decades, new approaches have been developed to assess the relation between real-time plant status and current environmental conditions which can be used for fast and accurate determination of the irrigation time [3-7].

The relationship between the temperature of canopy or leaf surface and water status of plants has been known for a long time [8]. It has been recognized as a potential tool for irrigation scheduling [4]. Estimating the water status through measuring the temperature of the canopy or leaf is based on the principle that the water lost through transpiration of the plant cools the leaves to below the temperature of the ambient air under well water conditions [1]. Studies conducted by Sdoodee and Kaewkong [9] and Blonquist et al. [10] showed that under water stress conditions, decreased water uptake induces stomatal closure or decreasing stomatal conductance leads to reduced transpiration and increases leaf temperature. Hackl et al. [6] also indicated highly significant relationships between canopy temperature and leaf water potential. Clawson and Blad [11] used remotely sensed techniques based on canopy minus air temperature (Tc-Ta) differential (?T), and suggested stress degree day (SDD) measure that is timely once midday (Tc-Ta) measured during the growth season. They proposed a crop water stress index (CWSI) to account for the effect of environmental factors especially the atmospheric vapor pressure deficit (VPD) and the net radiation (Rn). CWSI was based on the principle that there is an upper and lower limit for (Tc-Ta) in any given value of VPD [1,5,12]. Tc-Ta and VPD relationship is known as CWSI–base line. Raviv and Lieth [2] reported a linear relationship between Tc-Ta and VPD. In addition, VPD was also recognized as the function of air temperature and relative humidity. They found that in the stressed treatment, leaves were usually 2-3ºC warmer than that in the non-stressed treatment. Udompetailkul et al. [13] suggested that it might be appropriate to use air temperature and relative humidity as independent variables instead of VPD as a single variable.

Despite widespread efforts made during the last decades, the lack of a simple and precise predictor model that can describe the plant water requirement based on leaf-air temperature relations is felt. This study was conducted to find the relations and to construct a model which can estimate water demands (level of water stress) through meteorological factors (such as temperature and relative humidity) in different crops (with emphasizing on greenhouse tomato and cucumber) which ultimately led to designing and developing a hardware device to determine the plant water status precisely.

Materials and Methods

This study was conducted in the experimental greenhouse (soilless culture) and farm in Islamic Azad University, Isfahan branch, located in the east of Isfahan (latitude 32° 38¢ N, 51° 47¢ W, altitude:1517 m), during 2010 to 2015. Based on Köppen-Geiger climate classification, Isfahan is located in the class of dry (arid and semiarid) climates with warm and dry summers. The long-term average annual rainfall and temperature are 120 mm and 16°C, respectively. The greenhouses used in this study were arch shaped buildings which were 520 cm in height and had a polyethylene cover. Temperature and humidity were adjusted by warm air heating system, and evaporating cooling system. Night and day temperatures were adjusted to 19-21°C and 25-30°C, respectively, depending on factors like growth stage, light intensity, etc. Crops were planted in pots with a volume of 20 liters and at least five drainage holes in soilless media (45 percent perlite, 25 percent cocopeat and 25 percent peat) and then zeolite was added (5%). Crops were irrigated and fertilized by drip irrigation system (4 liters per hour discharge per plant) and fertigation method, respectively [2]. The open field soil texture was silty loam with EC=3.5 mmoh/cm, pH=7.8, and organic matters=0.8%.

For each greenhouse crops, 15 plants were sown (in 15 pots) and at daily intervals, five of them did not irrigate until water stress symptom (leaf wilt) was observed. Then, the moisture content of media (output or dependent variable) and other factors (input or independent variables, which are mentioned in each phase) were measured. In open-field crops, each one planted on 12 square meters with 2 planting dates (at 10 days intervals), 3 replications and two growing seasons. Similar to the greenhouse experiment, the variables were measured for the farm during one week, between two consecutive irrigations and at different hours in a day. Also, the wind speed was measured in the farm.

The experiment was performed in four phases as below:

First phase (2010-2011): determining the most efficient input variables for modeling crop water status

This phase was conducted on two greenhouse crops (cucumber and tomato) to find correlations between percentage of media culture moisture (PMM) (which is related to media culture water potential), VPD, photosynthetically active radiation (PAR), wet bulb temperature (WbT), Tl, Ta, Ta-Tl and relative humidity (RH) aiming at formulating the relation with simple mathematical model (equation). Ta,Tl, RH, and WbT were measured by Testo 625 with 0.1°C accuracy. PMM was measured by TDR 100 (America FIELDSCOUT Company) with a 1% accuracy. PAR was measured by Quantum Meter 1180 (µmol/m2/s) and VPD was calculated by the following formula:

equation (1)

The general procedure of the applied attribute selection technique is shown Figure 1. In the first step, 7 variables including VPD, PAR, WbT, Tl, Ta, RH and Ta-Tl were selected as the original set. The stepwise backward elimination method was employed in the subset generation and subset evaluation steps that include an iterative search for the optimal subset of variables. Starting with the full set of attributes, the procedure removes the worst attribute at each iteration, i.e., the removal of the attribute leads to the greatest increase in the determination coefficient (R) comparing with the previous iteration. For the subset evaluation, the subset of the input variables were fed to a multilayer feed forward neural network and its performance was measured for each subset. This type of neural network was chosen due to its accuracy and high tolerance of noisy data. The number of hidden layers and their units was chosen empirically. Once no increase is achieved for R-value at any iteration, the generated subset would be used to construct models for tomato and cucumber.

irrigation-and-drainage-systems-engineering-procedure

Figure 1: The overall procedure of the attribute selection method. ANN: Artificial Neural Network

Second phase (2011-2014): determining the model for other crops

In this phase, according to the results of the first phase, for determining irrigation time, only variables Tl, Ta, RH (the most relevant independent variables) and PMM were measured in the greenhouse and open-field crops within two growing seasons for each crop, to obtain the best model similar to first phase. Cucumber, tomato, strawberry and pepper were planted in the greenhouse while broad bean, wheat, barley, turnip, soybean, radish, melon, potato, maize and sugar beet were sown in the open-field.

In another experiment on cucumber and tomato in soilless culture, fruit yield in different irrigation regimes (based on PMM) was used as an index to determine the best irrigation time. These experiments were conducted in a completely randomized design with 4 replications in the greenhouse and 8 plants in each replication. Treatments included irrigation intervals based on moisture content percentage of 8, 10, 12, 15, 18 and 20 percent for cucumber and 5, 8, 10, 12, 15 and 18 percent for tomato. Media moisture content was measured by TDR. In addition, the plant daily yield (Kg) and the total yield during the growing season were obtained for each plot.

Overview of the prediction model

After determining the relevant variables, non-relevant attributes were eliminated from the original dataset and the remaining data were divided into training and testing sets. Furthermore, to obtain the taskrelevant data for the prediction, two pre-processing steps were applied on the raw data which help speed up the learning phase. In the first step, a few instances with missing values and outliers were identified and removed, and in the second step, the input features were normalized to the range of (-1, 1). This prevents attributes with initially large ranges from outweighing the attributes with initially smaller values. In the next step, the normalized inputs are fed into the network and the learning algorithm is applied until the whole network is converged.

Finally, the training and testing sets were used to construct an ANN prediction model and to evaluate the model, respectively. To determine the optimal architecture of the neural network, different number of nodes ranging from 2 to 30 were chosen for hidden layer and the resulting architecture was tested against the Akaike Information Criterion (AIC) [14] and determination coefficient. The AIC is calculated by the equation proposed by Burnham and Anderson 2003 [15].

AIC = 2k + n ln(RSS) (2)

in which k is the number of network weights, n represents the number of training tuples and the RSS is the residual sum of squares. In our case, the AIC provides a trade-off between the performance of the neural network in the training phase and the size of hidden layer such that the lower values of AIC represent the goodness of the model. It may be seen that, by increasing the number of nodes, a descending trend is observed for the AIC.

As shown in Figure 2, the determination coefficient increases with increasing number of neurons until the number of neurons reaches 14, and then it starts to decrease slowly thereafter. This implies that overfitting is occurred after the network is sufficiently trained and an optimum value is obtained for the number of nodes in the hidden layer. In addition, small vibrations with no significant change is noticed for the AIC beyond this point. Therefore, the optimal number of neurons for the hidden layer was chosen as 14.

irrigation-and-drainage-systems-engineering-training

Figure 2: The effect of number of nodes on the performance of the model in the training phase.

A variety of transfer function combinations for hidden and output layers were studied. The results showed that tangent-sigmoid (tansig) and linear (purelin) functions led to the best performance in terms of determination coefficient which reflects the relationship between the outputs of the network and the targets.

The Levenberg-Marquardt back-propagation learning algorithm was employed for training the model and half of the dataset instances were randomly selected to construct the training set. The remaining instances were divided into two partitions representing validation and test sets. The learning rate and the training goal were selected as 0.01 and 0.05, respectively and the maximum number of passes through the set of training instances along with the associated updating of the weights (epochs) was set to 30. The Mean Squared Error (MSE) was used to validate the learning process through 10-fold validation technique.

Once the network is trained, a model is obtained by combining the weight matrices between adjacent layers of the neural network. Considering the basic components of a node shown in Figure 3, the process of extracting the prediction model can be presented as follows:

irrigation-and-drainage-systems-engineering-artificial

Figure 3: Basic elements of a node in an artificial neural network.

Let X=[x1 x2…xn] be a vector of input values and Wh be a n × m matrix representing the weights of a fully connected network between n nodes in input layer and m nodes in the hidden layer. The total input T entering the hidden layer is calculated by

equation (3)

where bh is a vector of size m representing the values of bias component of the nodes in the hidden layer. As shown in Figure 3, each node is associated with a correction weight with a constant non-zero value called bias. It serves as a thresholding variable to control the output of the transfer function in the corresponding node. It should be noted that, the learning algorithm also updates the bias vector to achieve optimal bias values in the final iteration of the algorithm.

Let the transfer functions employed in the hidden and output layers are denoted by fh and fo. The second step involves applying the activation function fh to the total input T entering the hidden layer which returns the output of the hidden layer Yh:

equation (4)

In the proposed prediction model, the tangent-sigmoid transfer function is used for fh. Assuming that the input of node i in the hidden layer is denoted by Ti, the output of the node is given by:

equation (5)

In the above formula, the superscript i in notation equation indicates the index of the node. The final output Y of the network is computed by propagating Yh through the connections between the hidden and output layers using similar formula shown in equation (3):

equation (6)

where bo is the bias of the output node used to calculate to final output of the neural network and Wo denotes the vector of weights between nodes in the hidden layer and the output node. As mentioned earlier, a linear transfer function is employed for fo, so that, the final output of the network can be represented as:

equation (7)

where m is the number of neurons in the hidden layer. The constructed model is used to predict the crop water stress which is a continuous numerical variable. This value can be discretized to obtain nominal values {Low, Medium, and High} for determining the crop water requirement for a single type of plant. When a sufficient number of observations is collected, the training data can be generalized to a wider category of plants and various irrigation conditions.

Third phase (2014): device design (Irrigation Time Detection System: ITDS)

The hardware of ITDS was implemented in the form of two separate sub-systems including information collection system and analyzer system. Information collection system is an AVR-based microcontroller system equipped with three sensors MLX90614 infra-red sensor to determine Tl, Sensirion_Humidity_SHT11 sensor to detect Ta and RH, and HC-03/05 Embedded Bluetooth Serial Communication Module which sends the sensed data to a cell phone. The software of this system is designed and programmed in Code Vision Integrated Development Environment (IDE). Also, Two Wire Interface (TWI) protocol is used for communication between microcontroller and the sensors. In this system, collected information is sent by the microcontroller in response to the user request through Bluetooth port to the analyzer part (i.e. the second part of ITDS implemented on a cell phone).

Fourth phase (2015): testing ITDS performance

In order to check the accuracy of ITDS, an experiment was conducted on greenhouse tomato. In this experiment, different media moisture status for tomato was determined by TDR and ITDS simultaneously and t-test is used to examine if the data obtained from these two methods were statistically correlated.

Results and Discussion

First phase: determining the most efficient input variables for modeling crop water status

High R-values (more than 0.7) obtained for both neural network models indicate that the independent variables used in this study (Table 1) are highly relevant to the prediction of PMM. The results shown in Table 1 indicate that no important changes is observed for coefficient of determination (R) when VPD, PAR, WbT, and Tl are removed from the datasets of both cucumber and tomato. Similar results were obtained in the research conducted by Udompetailkul et al. [13], where leaf temperature, air temperature and humidity played important roles in the estimation of water demand for almond trees. Their study showed that VPD and WbT are correlated with Ta and RH, respectively. Also, PAR is directly affected by Ta, Tl, and RH. Therefore, the values of Ta, RH, and Ta-Tl have a major effect on the estimation of VPD, PAR, WbT and Tl.

No PAR VPD WbT RH TA TL TA-TL Rave(Tomato) Rave(Cucumber)
1 + + + + + + + 0.83 0.81
2 - + + + + + + 0.8 0.78
3 + - + + + + + 0.79 0.82
4 + + - + + + + 0.82 0.8
5 + + + + + - + 0.82 0.83
6 - - + + + + + 0.86 0.79
7 - + - + + + + 0.81 0.78
8 - + + + + - + 0.83 0.78
9 + - - + + + + 0.8 0.81
10 + - + + + - + 0.78 0.8
11 + + - + + - + 0.81 0.78
12 - - - + + + + 0.81 0.81
13 - - + + + - + 0.8 0.8
14 + - - + + - + 0.79 0.82
15 - + - + + - + 0.81 0.78
16* - - - + + - + 0.81 0.8

Table 1: Average of coefficient of determination (R) for different combination of attributes.

The last row of Table 1 implies that, even though VPD, PAR, WbT and Tl variables were removed from the model, an appropriate regression model was obtained without significant change in the performance of the model. For example, in the second row of Table 1, by eliminating PAR variable, R was obtained as 0.80 and 0.78 for tomato and cucumber, respectively. As a part of our experiment, PAR was used as a dependent variable and efficient variables (Ta, Ta-Tl, and RH) were simulated as independent variables by neural network. The results showed a high correlation between these factors and PAR (R=0.83). When two independent variables are highly correlated, one of them should be removed to avoid redundancy and overfitting problems. This is the reason that removing PAR from the model did not make a remarkable change in R, despite its importance.

Previous studies also proved that VPD is related to Tc-Ta [3,13,16-19]. On the other hand, based on the formula of VPD denoted in equation (1), Ta and RH have the direct effect on VPD value. Similarly, in our experiments, after elimination of VPD variable from the model, R was obtained 0.79 and 0.82 in tomato and cucumber respectively. Also, Wanjura et al. [7] proved that a canopy is likely to be cool only about 2°C above the ambient wet bulb temperature. According to Table 1, R-value is not affected by the elimination of WbT from the model. The results showed that omission of leaf temperature has a trivial impact on R (R=0.83) because Ta-Tl exists in the model. Therefore, 3 variables Ta, Ta-Tl and RH were selected as inputs of the neural network in the second phase to obtain more precise and simple models.

Second phase: determining the model for other crops

Incorporating the ranges of input and output variables and their normalization coefficients shown in Table 2, the values of Ti (i=1,2,…,14) are given by:

Variable/Unit Minimum Maximum Normalization Coefficient (Gain)
 Ta (°C) 19.2 39.7 0.0976
 RH (%) 8.2 75.6 0.0297
Ta-Tt (°C) 0.3 18.8 0.1081
 PMM (%) 2.4 27.17 0.0807

Table 2: Ranges and normalization coefficients of input and output variables.

equation

equation

Based on the results of the first phase, in the second phase, MSE and R value were obtained only with 3 variables (Ta, Ta-Tl, and RH). The results obtained from 4 greenhouse crops and 10 open field crops are presented in Table 3. As the training of the neural network was completed and the convergence conditions were met, the lowest mean square error was obtained for the three categories of data (i.e., data used in the training, validation, and testing steps), and the amount of deviation from the mean (error) was calculated accordingly. These results showed that the moisture content in media was related to three efficient variables (Ta, Ta-Tl, and RH), evapotranspiration and stomata conductance of plants. Stomata conductance depends on Ta, PAR, and RH. The results of the study conducted by Çamoglu [20] also showed that the relationship between Ta-Tl and evapotranspiration was linear, and the coefficients of determination were statistically significant in two cultivars of olive. Olufayo et al. [21] suggested that Mid-day measurement of Ta-Tc reached to the maximum of 7°C in the dry treatments and was maintained close to 0°C in full irrigated treatments.

  Plant species parameters in neural network Number of samples MSE R (all)
Greenhouse Cucumber Training 2327 6.73 0.88  
(Cucumissativus) Testing 499 5.92 0.87 0.87
Tomato Training 1651 7.17 0.82  
(Solanumlycopersicum) Testing 354 7.66 0.81 0.82
Strawberry Training 549 25.33 0.8  
(Fragaria × ananassa) Testing 117 32.88 0.7 0.78
Pepper Training 737 66.24 0.84  
(Capsicum annuum) Testing 158 83.05 0.8 0.85
Open Field BroadBean
(Viciafaba)
Training 356 2.27 0.93 0.98
Testing 76 2.2 0.99
Wheat
(Triticumaestivum)
Training 352 40.18 0.81 0.81
Testing 75 56.96 0.8
Barley
(Hordeumvulgare L.)
Training 346 50.08 0.86 0.82
Testing 74 63.45 0.84
Turnip
(Brassicarapasubsp. rapa)
Training 336 64.47 0.85 0.85
Testing 72 56.03 0.89
Soybean
(Glycine max)
Training 361 9.91 0.93 0.86
Testing 78 19.67 0.87
Radish
(Raphanussativus)
Training 348 22.73 0.86 0.86
Testing 74 72.68 0.79
Muskmelon
(Cucumismelo)
Training 363 4.66 0.89 0.86
Testing 78 7.25 0.89
Potato
(Solanumtuberosum L.)
Training 347 7.81 0.86 0.84
Testing 74 40.01 0.8
Maize
(Zea mays subsp. mays L.).
Training 361 10.4 0.85 0.85
Testing 77 15.14 0.86
Sugar beet
(Beta vulgaris)
Training 352 33.04 0.73 0.73
Testing 76 37.98 0.75

Table 3: The results of applying the neural network for predicting crop water requirement for greenhouse and open field plants in the second phase.

In another research, multiple linear regression models of leaf temperature as functions of stem water potential, air temperature, relative humidity, photosynthetically active radiation, and wind speed were developed and validated for almond and walnut crops under sunlit and shaded conditions. Models yielded high correlation with R-values ranging from 0.82 to 0.90. Discriminant analyzes the data obtained from the sensor suite resulted in error rates of 9% to 11% in walnuts and 16% to 17% in almonds [19]. Wang et al. [22] in another study measured midday canopy to air temperature differences in the water-stressed postharvest deficit irrigation treatments in the 5-7°C range, which were consistently higher than the 1.4-2°C range found in the non-water-stressed control treatments. A reasonable correlation (R=0.67-0.70) was obtained between stem water potential and the canopy to air temperature difference, indicating the possibility of using the canopy temperature to trigger irrigation events.

According to the analysis of variance and mean fruit yield comparison between different levels of media moisture, the following results were obtained in greenhouse cucumber and tomato (Table 4):

Tomato   Cucumber  
W Yield (Kg) W Yield (kg)
18% 8.45 20% 21.72
15% 8.34 18% 19.85
12% 6.96 15% 16.83
10% 6.70 12% 14.45
8% 5.40 10% 13.17
5% 5.65 8% 11.51

Table 4: Mean comparison between different media moisture (W) in cucumber and tomato.

The results of the study showed that maximum yield of cucumber and tomato were attained in 18% and 15% of the media moisture, respectively. These observations showed that different crops had different thresholds of media moisture for irrigation time, which should be programmed in the device designed for prediction of precise irrigation time. Wanjura et al. [7] suggested a method to determine a temperature threshold for controlling drip irrigation scheduling of cotton. They showed that maximum yields were produced when the threshold is selected between 28°C and 30°C for canopy temperatures.

Third phase: device design (Irrigation Time Detection System: ITDS) and implementation of ITDS system

Considering the required functionalities to implement the data analysis system such as low processing load and CPU frequency, we found that the ITDS system could be embedded in particular cell phone hardware. As illustrated in Figure 4, the software of this system comprises of three modules including data collection part, Bluetooth connection interface and an ANN-based data analysis module. The software was implemented and tested in Eclipse Integrated Development Environment based on Android APIs.

irrigation-and-drainage-systems-engineering-cellphone

Figure 4: Data collection and cellphone with data analysis software of Irrigation Time Detection System (ITDS).

In the data analysis module, the neural network parameters including input and output weights, normalization coefficients, number of nodes, transfer functions and bias values are read from a local database and are used to initialize the ANN software module. The local database is used to store the mentioned parameters and water stress thresholds for different types of crops. When the user changes the plant type, data analysis software reads the ANN parameters for the corresponding plant. Then, the input variables (Ta, Tl, and RH) are provided by the sensors and fed to the data analysis module to efficiently predict the water demand (media moisture) of the plant. Finally, ITDS displays the water demand status to the user.

Fourth Phase: testing ITDS performance

The results of ITDS test are presented in Table 5. According to the results, the difference between the means of predicted values calculated by ITDS and TDR readings were very low (0.08%) and the mean difference between them was not significant (P value=0.85). It implies that ITDS could simply predict the media moisture with high accuracy.

Average ITDS read Average TDR read T value P value
15.73 15.81 0.19 0.85

Table 5: Results of paired – T test for comparing the ITDS and TDR values.

Conclusion

In this paper, a mathematical model based on multilayer feed forward neural network is proposed to predict the media moisture from Ta, Ta-Tl and RH variables. Due to its simplicity, the proposed model can be efficiently implemented in the embedded systems using low level programming languages. In this study, a complete irrigation time detection system (ITDS) is introduced in which the proposed model can be applied for different types of crops and plant species. ITDS was designed to predict water demands simply by sensing air temperature, canopy temperature and relative humidity which facilitates its use among experts and non-expert users. Based on our experiences obtained during the ITDS tests, the usage of ITDS is not suggested in the following situations:

• When the wind speed is too low (causing a boundary layer on the surface of the leaf).

• When the wind speed is too high.

• When the temperature is lower than the demand of the plant.

• When the temperature is too high (causing midday stress).

• When transpiration is too low, i.e. the relative humidity is too high or the light intensity is too low; for example in early morning after sunrise and before sunset.

• In sudden changes of the weather; for example sunny to cloudy.

The above limitations can be considered as the topics for further research in this field.

Acknowledgements

This study was supported by Isfahan (Khorasgan) Branch of Islamic Azad University and agricultural organization of Isfahan province. We appreciate the Board of Directors for this financial support.

References

Select your language of interest to view the total content in your interested language
Post your comment

Share This Article

Recommended Conferences

Article Usage

  • Total views: 73
  • [From(publication date):
    March-2017 - Apr 29, 2017]
  • Breakdown by view type
  • HTML page views : 55
  • PDF downloads :18
 
 

Post your comment

captcha   Reload  Can't read the image? click here to refresh

Peer Reviewed Journals
 
Make the best use of Scientific Research and information from our 700 + peer reviewed, Open Access Journals
International Conferences 2017-18
 
Meet Inspiring Speakers and Experts at our 3000+ Global Annual Meetings

Contact Us

Agri, Food, Aqua and Veterinary Science Journals

Dr. Krish

agrifoodaquavet@omicsonline.com

1-702-714-7001 Extn: 9040

Clinical and Biochemistry Journals

Datta A

clinical_biochem@omicsonline.com

1-702-714-7001Extn: 9037

Business & Management Journals

Ronald

business@omicsonline.com

1-702-714-7001Extn: 9042

Chemical Engineering and Chemistry Journals

Gabriel Shaw

chemicaleng_chemistry@omicsonline.com

1-702-714-7001 Extn: 9040

Earth & Environmental Sciences

Katie Wilson

environmentalsci@omicsonline.com

1-702-714-7001Extn: 9042

Engineering Journals

James Franklin

engineering@omicsonline.com

1-702-714-7001Extn: 9042

General Science and Health care Journals

Andrea Jason

generalsci_healthcare@omicsonline.com

1-702-714-7001Extn: 9043

Genetics and Molecular Biology Journals

Anna Melissa

genetics_molbio@omicsonline.com

1-702-714-7001 Extn: 9006

Immunology & Microbiology Journals

David Gorantl

immuno_microbio@omicsonline.com

1-702-714-7001Extn: 9014

Informatics Journals

Stephanie Skinner

omics@omicsonline.com

1-702-714-7001Extn: 9039

Material Sciences Journals

Rachle Green

materialsci@omicsonline.com

1-702-714-7001Extn: 9039

Mathematics and Physics Journals

Jim Willison

mathematics_physics@omicsonline.com

1-702-714-7001 Extn: 9042

Medical Journals

Nimmi Anna

medical@omicsonline.com

1-702-714-7001 Extn: 9038

Neuroscience & Psychology Journals

Nathan T

neuro_psychology@omicsonline.com

1-702-714-7001Extn: 9041

Pharmaceutical Sciences Journals

John Behannon

pharma@omicsonline.com

1-702-714-7001Extn: 9007

Social & Political Science Journals

Steve Harry

social_politicalsci@omicsonline.com

1-702-714-7001 Extn: 9042

 
© 2008-2017 OMICS International - Open Access Publisher. Best viewed in Mozilla Firefox | Google Chrome | Above IE 7.0 version