Comparibson Between two Classification Methods of Maximum likelihood and Artificial Neural Network for Providing Land use Maps Case Study: Ilam Dam Area
One of the most necessary information required by the managers and custodians of natural resources are land use maps. Satellite data, for owning characteristics such as presenting on time and digit information, variable forms and process ability have a great role in providing land use maps. In the o...
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | fas |
| Published: |
University of Sistan and Baluchestan
2010-12-01
|
| Series: | جغرافیا و توسعه |
| Subjects: | |
| Online Access: | https://gdij.usb.ac.ir/article_633_c687c05bd134032ea5f20c9fdd5df24b.pdf |
| Summary: | One of the most necessary information required by the managers and custodians of natural resources are land use maps. Satellite data, for owning characteristics such as presenting on time and digit information, variable forms and process ability have a great role in providing land use maps. In the other hand, during the recent years, advanced classification methods including artificial neural networks, fuzzy sets and intelligent systems are widely used for classification of satellite photos. The main objective of this research would be comparing the two different methods of classification for land use by the use of ASTER photos. For this reason, by using ASTER satellite photos and two supervised classified algorithms including maximum likelihood and artificial neural network, the land use map was prepared. In classification with neural network algorithm, a Perceptron network with a hidden layer, 14 input neurons ‘9 middle neurons and 6 output neurons have been used for classification by neural network algorithm, in which the number of input neurons are the same number of ASTER satellite photo bands and the number of output neurons are the same number of classes for land use map. For network training, back propagation algorithm has been used.
The results obtained from accuracy evaluation of these two methods by the use of Kappa coefficient showed that neural network algorithm with coefficient 0.86 in comparing with maximum likelihood algorithm with coefficient 0.69 was more accurate. The results of this study show that traditional classification algorithms like statistical methods for its low flexibility and its different parametric types like maximum likelihood methods for their depending on Gaussian model cannot provide optimized results in case of abnormality of educational data, while the reason of success of artificial neural network algorithm in remote sensing is that it is able to integrate data with different resources. |
|---|---|
| ISSN: | 1735-0735 2676-7791 |