Comparison of different methods of classification of satellite images in preparing land use map (Case study: Lake Urmia)

Subject Areas :

Hossein nazmfar ¹ , Monir Shirzad ²

1 - عضو هیئت علمی دانشگاه محقق اردبیلی
2 - University of Mohaghegh Ardabili

Received: 2022-02-04 Accepted : 2022-02-23 Published : 2022-02-23

Keywords: Classification, land use, Satellite image, Urmia Lake, Supervised Methods,

Abstract :

The purpose of this study is to compare nine supervised methods in classifying land cover using Landsat 8 satellite images in Urmia Lake. The nature of this research has been developmental-applied and the method of performing it has been descriptive-analytical. For this purpose, satellite images of OLI sensor related to the date (7/8/2016 and 7/6/2016) were downloaded from the USGS site. And after applying the pre-processing using field visits and the Global Positioning Machine, instructional samples were prepared for each user (Lake, Agricultural land, Salty land, Waste land) in the study area. In the next step, the classification was performed using nine monitored algorithms (neural network, backup vector machine, maximum probability, mahalanobis, minimum distance from average, parallel surfaces, spectral information divergence, binary codes, spectral angle). In the last step, in order to check the accuracy and precision of image classification, evaluation criteria (manufacturer's accuracy, user accuracy, overall accuracy, kappa coefficient) were used. The results indicate that the classification method of backup vector machine with 99.57% capa coefficient after neural network vector support vector machine with 98.66% cappa coefficient and the maximum probability method with 98.58% capa coefficient after neural network method compared to other methods They are more accurate. Also the least accurate are binary code algorithms with parallel surfaces and spectral angles.