XML Persian Abstract Print

university of tabriz , somayeh.mehrabadi@yahoo.com
Abstract:   (1088 Views)
Modelling forest degradation using mathematical approaches is a difficult undertaking. Accordingly, forest degradation modeling using old-fashioned methods (hard computing) is very difficult, if not impossible; because such methods are based on accuracy of variables and calculations, but our real world phenomena operate based on indistinct, indeterminate borders. Such indeterminacy is consistent with soft computing methods.  Among such approaches are neural network and fuzzy inference system, each has its own drawbacks if it is used alone. In order to overcome the drawbacks, hybridization has been proposed in which two or more intelligent methods are combined with the aim of eliminating the drawbacks of each of those single methods. In this study, forest degradation was modelled by employing single perceptron neural network and neuro-fuzzy combined methods. As a result, were utilized images from Landsat-5 TM sensor for 1999 and Landsat 8 OLI sensor for 2017. The output of analysis was the images of forest and non-forest areas for the two years. Next, were sampled the degraded and non-degraded areas. Taking into consideration the previous studies and analysis results, were identified 7 factors as the most effective agents in forest degradation. (distance for (city-river-village-sea-road) and elevation & slop). After that, were modeled the Perceptron networks and the adaptive neuro-fuzzy network with the measured variables for 200 locations. In Perceptron approach, the data was trained within the neural network. Furthermore, mean square error was used for evaluating the performance. The MSEs for the perceptron neural network using three algorithms of Levenberg Marquardt, Bayesian Regularization and Scaled Conjugate Gradient were ,0.053, 0.070,   and 0.090, respectively. On the other hand, the MSEs for neuro-fuzzy model using optimization and combined methods were 0.019 and 0.0102, respectively. However, in adaptive neuro-fuzzy approach, besides using the learning ability of Artificial Intelligence (AI), were employed fuzzification as well. Finally, the results of models and the error rates of both methods indicated the relatively desirable performance of soft computing methods. The adaptive neuro-fuzzy model, the latter has a better performance and is closer to the reality due to its reliance on indeterminacy.
Type of Study: Research | Subject: Special
Received: 2018/09/10 | Accepted: 2019/04/18

Add your comments about this article : Your username or Email:

Send email to the article author

© 2019 All Rights Reserved | Scientific Journals Management System

Designed & Developed by : Yektaweb