Performance comparison between Multi layer perceptron and Radial Basis Function networks to predict commercial and cordwood volume of trees

Document Type : Complete scientific research article

Authors

Abstract

Background and objectives: In forest resource management, decision-making processes, such as qualitative factors, are not logged in mathematical equations so we need to new solutions than algorithmic methods. According to the capabilities of neural networks and recent application of them in forest resources, the purpose of this study was to compare the multi layer perceptron and the radial basis network to predict commercial and cordwood volume, in order to evaluate the performance of different networks to find the best type of network for achieving acceptable and valid results.
Materials and methods: In this purpose, 367 trees were marked of research and educational forest of kheyroud. Some factors such as diameter at breast height, diameter at stump, stump height, total height, topographic factors (slope, aspect and elevation), species, tree situation and minimum median diameter of last log were selected and then they were measured. They considered as input variables in network. Commercial and cordwood volume determined by traditional renewal volume and then they used as output network. Multi-layer perceptron (MLP) and radial basis function (RBF) were used for modeling. The hyperbolic tangent function and softmax function respectively used for network training in hidden layer of multi layer perceptron and radial basis function networks. As well as, the linear function used for network training in output layer. The data were divided into three sections for modeling: training, validation and test, each of which was 70%, 15% and 15%, respectively. Determination of the number of hidden layers and neurons of each layer was also performed by test and error and continued until the error value reached the minimum.
Result: Due to result, R2 value was respectively 0.94, 0.71 for commercial and cordwood volume in multi-layer perceptron network and 0.88, 0.65 for cordwood volume in radial basis function network. Also, RMSE value was respectively 1.297, 0.337 for commercial and cordwood volume in MLP network and 3.72, 0.397 for cordwood volume in RBF network.
Conclusion: The result showed that multi-layer perceptron than radial basis network has acceptable accuracy to predict the commercial volume and cordwood volume. The only advantage of the radial basis function than multi-layer perceptron was less time required for training in modeling. Using a network and a model that has a higher accuracy with several variables among existing networks and models is prioritized. Thus, according to this new and powerful technique, the need for identifying a range of potential uses in the forest science community is felt as an alternative tool.

Keywords

Main Subjects

References

1.Bayat, M., Namiranian, M., Omid, M., Rashidi, A., and Babaei, S. 2016. Applicability of artificial neural network for estimating the forest growing stock. Iranian Journal of Forest and Poplar Research. 24: 2. 214-226. (In Persian)
2.Bayati, H., and Najafi, A. 2011. Application of artificial intelligence in trees stems volume estimation. Journal of Renewable Natural Resources Research. 2: 2. 52-59. (In Persian)
3.Bayati, H., Najafi, A., and Abdolmaleki, P. 2013. Comparison between Artificial Neural Network (ANN) and Regression Analysis in Tree Felling Time Estimation. Iranian Journal of Forest and Poplar Research, 20: 4. 595-607. (In Persian)
4.Coulson, R.N., Folse, J.L., Loh, D.K. 1987. Artificial intelligence and natural resource management. Science. 237: 262-267.
5.Diamantopoulou, M.J. 2005. Artificial neural networks as an alternative tool in pine bark volume estimation. Computers and Electronics in Agriculture. 48: 235–244.
6.Diamantopoulou, M.J. 2006. Tree-Bole Volume Estimation on Standing Pine Trees Using Cascade Correlation Artificial Neural Network Models. Agricultural Engineering International Manuscript IT. 06 002: 1-14.
7.Diamantopoulou, M.J., and Milio, E. 2010. Modelling total volume of dominant pine trees in reforestations via multivariate analysis and artificial neural network models. Biosystem engineering. 105: 306-315.
8.Gimblett, R.H., and Ball, G.L. 1995. Neural network architectures for monitoring and simulating changes in forest resources management. AI Applications. 9: 2.103-123.
9.Gorzin, F. 2015. Prediction volume of trees by artificial neural networks (Case Study: kheyroud Forest). M.Sc. thesis, Faculty of Natural Resources, University of Tehran, Karaj, 78p. (In Persian)
10.Kia, M. 2010. Neural Network in Matlab. Kian Rayaneh Sabz Publisher, Tehran, 323p. (In Persian)
11.Ozçelik, R., Diamantopoulou, J.M., Brooks, J.R., and Wiant Jr, H.V. 2010. Estimating tree bole volume using artificial neural network models for four species in Turkey. Journal of Environmental Management. 91: 742–753.
12.Peng, C., and Wen, X. 1999. Recent Applications of Artificial Neural Networks in Forest Resource Management: An Overview, Environmental Decision Support Systems and Artificial Intelligence. 15-22.
13.Safi Samgh Abadi, A. 2003. Forest multi-objective planning by artificial neural networks. Ph.D. thesis, Faculty of Natural Resources, University of Tarbiat Modarres. Noor, 156p. (In Persian)
14.Soltani, S., Sardari, S., Sheykhpour, M., and Mousavi, S.S. 2010. Introduction to fundamentals and Artificial Neural Network applications. NS Scientific and Cultural institute. Tehran. 216p. (In Persian)
15.Vahedi, A., Mataji, A., and Akhavan, R. 2017. Modeling the commercial volume of trees in mixed beech stands of Hyrcanian forests through artificial neural network. Forest and Wood Products. 70: 1. 49-60. (In Persian)
Journal of Wood and Forest Science and Technology
Volume 24, Issue 4
March 2018
Pages 187-198

Files

Share

How to cite

Statistics