Agricultural Research Management Information System

Home
Research Summary
Search
About Us
- ARMIS
- Brochure
Contact Us
Report
User Request
Data Input
Help
- Operation Manual
  - PDF
  - Video
- Program Area & Commodity
We have reached 37600 number of research entries at this moment.

- Logout

Research Detail

Home
Research
Detail

Modeling Jute Growth Using a New Technique of Artificial Neural Networks: Model Development and Description

Dr. Md. Mujibur Rahman
CSO
Jute Farming Systems Division Bangladesh Jute Research Institute Manik Mia Avenue Dhaka-1207

Abstract/ Executive Summary:

Neural network models, trained by back propagation, were developed to predict the development of jute plant using previously obtained experimental data. The models were based on a common structure and devolved using data from six sets of experiments conducted in 2006 and 2007 on two different varieties of jute. The models consisted of four-layered networks and a large number of neurons. The six input variables were represented by six neurons; julian day, solar radiation, maximum temperature, minimum temperature, rainfall, and type of biomass or photosynthetic leaf area. The output variable, represented by a single neuron, was plant dry matter or photosynthetic leaf area. The models had two hidden layers with 9 and 5 neurons. Three sets of experiments conducted in 2006 were used for training the models and another three sets of experiments conducted in 2007 used for models validation.

Keywords: Modeling jute growth, Neural networks, Jute

Location: Central Research Station (CRS) of Bangladesh Jute Research Institute, Jagir, Manikgonj,

Start Date: 00-00-2006

End Date: 00-00-2007

Program Area: Knowledge Management

Commodity: Jute

Objectives:

To artificial neural networks (ANNs) as an alternative modeling approach for predicting the jute growth.

Methodology:

Field experiments were conducted in the Central Research Station (CRS) of Bangladesh Jute Research Institute, Jagir, Manikgonj, Bangladesh to determine biomass of the different components of the jute plant during its growth in development stages. Data on Plant Height (PH, cm), plant Base Diameter (BD, mm), Leaf Area development (LA, cm2), Leaf Area Index (LAI), plant Total Green Weight (TGW, g plant-1), and green and dry matter weights (g plant-1) for Root (RGW and RDM), Bark (BGW and BDM), Stick (SGW and SDM) and Leaves (LGW and LDM) as well as the plant Total Dry Matter weights (TDM, g plant-1) were collected at an interval of 10 days starting from the plant age of 20 DAE (days after emergence). Also final weights of jute fibre (kg plot-1) and jute sticks (kg plot-1) were recorded. Six sets of field experiments were conducted in jute growing seasons in 2006 and 2007. Six sets of experimental data for two varieties of jute Corchorus capsularis L (cv. CVL-1) and C. olitorius L (cv. O-9897) were collected based on split plot experimental design with variety in main plot and date of harvest (data collection) in split plot with 3 replications. Jute plant dry matter changes during the growth stages (DAE) of the year 2006 were used for the training purpose of the ANN models and other data sets for the year 2007 were used to test the ANN models for jute plant growth. The most popular neural-network paradigm is the backpropagation learning algorithm. The back-propagation neural network has been applied with great success to model many phenomena in the field of yield prediction and drying. Each hidden and output neuron processes its input(s) by multiplying each by its weight, summing the product, and then processing the sum using a nonlinear transfer function (activation function) to obtain the desired result. The most common transfer function implemented in the literature is the sigmoid function. The neural network “learns” by modifying the weights of the neurons in response to the errors between the actual output values and the target output values. Training is carried out by repeatedly presenting the entire set of training patterns (updating the weights at the end of each epoch) until the average sum squared error over all the training patterns is minimal and within the tolerance specified for the problem. At the end of the training phase, the neural network should correctly reproduce the target output values for the training data; provided errors are minimal (i.e. convergence occurs). The associated trained weights of the neurons are then stored in the neural network memory. In the next phase, the trained neural network is fed a separate set of data. In this testing phase, the neural network predictions using the trained weights are compared to the target output values.

Source of Information: Bangladesh J. Jute Fib. Res. 2013, 30 (1-5): 99-110

Article Link (Online Journal):

Funding Source:

Total Budget (Tk.) :

Achievement/Findings:

Experimental studies on jute production systems for two varieties of jute C. capsularis L (cv. CVL-1) and C. olitorius L (cv. O-9897) were conducted and there found no significant difference for the growth patterns between these two varieties of commercially important jute. Artificial neural network models were developed to predict the growth patterns of the different components (dry-matters and photosynthetic leaf area) of the jute plant and it was programmed in C++. ANN models predicted the growth of the different components of jute plant accurately. Predicted results are presented in the article “Modeling Jute Production System using a New Technique of Artificial Neural Networks-II: Model Simulation, Validation and Application” in this series.

Knowledge Management: Journal