Introducing an intelligent algorithm for extraction of sand dunes from Landsat satellite imagery in terrestrial and coastal environments

In this study, an intelligent algorithm for the extraction of sand dune maps from Landsat satellite images in terrestrial and coastal environments was presented by a case study of the sand dunes in northeastern Iran and one example on the coast of the Oman Sea in southeastern Iran. First, to find the best sand dune detection algorithm, the land cover map is created by aerial photography and field observation in ILWIS software. Next, a binary image is created in which the sand dunes are considered as white pixels and other parts as black pixels. In the next step, the pixels of each land cover map class are divided into 50 equal parts, one part of them is used to train different smart grids and the rest of them are used for evaluation. Then the classification is done using all 50 parts in turn, and the Correct Classification Rate (CCR) index is calculated. This action is performed for the K-Nearest Neighbor, Decision Tree, AdaBoost, RUSBoost, and SVM algorithms. Furthermore, to see the effect of the PCA and LDA statistical methods, the classifier algorithms are implemented by applying these techniques. Finally, the proposed method was tested in a coastal area as an example. The findings show that the CCR of the proposed method is around 84% using 2% of the ground truth of the area for training and the other 98% of them for testing. In the case study, the detection results show that around 9.98% of the area is covered by sand dunes. To have a practical detection of the sand dunes area, the proposed machine learning method is applied to some experimentally verified points. Based on the results, ensemble classification methods, especially RUSBoost, have the best detection accuracy. Also, the result of its test on the coastal sand dunes showed its ability to identify, map, and monitor the sand dunes in coastal areas.