عنوان مقاله [English]
Background and Objectives:
Soil salinity is one of the major environmental problems affecting vast areas in many countries and this problem reduces the ability to produce and fertilize the soil for cost-effective production. Identification and monitoring of saline areas is necessary for controlling land degradation behavior and its sustainable management, especially in semi-arid areas. Expansion of the process of soil salinization is one of the important environmental challenges of the eastern margin of Urmia Lake. The research results of the researcher indicate that less work has been done in this salinity field by means of object-oriented methods. As well as among the various types of classifiers in the eCognition software environment, comparisons between salient classification algorithms are less significant. Therefore, it is necessary to compare the images classification algorithms and algorithms that have a good accuracy in extracting image complications are identified.
Materials and Methods:
The study area is part of the eastern catchment of Urmia Lake which includes the plains of Tabriz, Shabestar, Oskuo, Azarshahr, Ajbashir and Bonab, Malekan and part of Maragheh. And it is located in the range of 37 degrees, 9 minutes to 38 degrees, 11 minutes north latitude and 45 degrees, 41 minutes to 46 degrees and 17 minutes east longitude in the northwest of Iran. The study area is 3847 square kilometers. For this purpose, two types of data including satellite images and GPS data obtained from field operation were employed. First, the stages of image preprocessing, including radiometric correction, to calculate the spectral indices, subset, and image bands stack in the ENVI 5.1 software were performed. After performing the preprocessing stages, the images and GIS data include (Topographic data from DEM 30m of area, soil classification, vegetation index (NDVI) and other information layers) were imported to the eCognition software for object based image processing and to classify knowledge based classification based on different algorithms. In this study, the effectiveness of the object based fuzzy technique and the nearest neighbor approach in the extraction of saline areas in the eastern of Urmia Lake were also investigated as part of methodology.
In order to achieve high accuracy results, using the ESP algorithm, the segmentation scale was optimized. By predicting the appropriate scale for creating image units using the ESP algorithm, the scale 170 with coefficients of shape 0.2 and compression 0.5 was selected as the appropriate scale for extraction of image symptoms. Accordingly, to classify the nearest neighbor classification, at the first, the intervals between educational samples were optimized using FSO algorithm. And for the image of the study, the eighteenth dimension with a separation distance 2.52 was identified as the best dimension for the separation of the desired classes. The results indicate that both methods with small difference have relatively similar results. However, the nearest neighbor approach has been estimated non-saline lands more than the object based fuzzy method. Based on our statement it could be due to the presence of mixed pixels
The results showed that the fuzzy object-oriented method, due to the use of membership degree functions, has overall accuracy of 0.94 and a kappa coefficient of 0.91, and it is superior to the nearest neighboring algorithm in extraction of eastern lake margins. It was also found that the brightness index is the most effective indicator in identifying and distinguishing salty lands from non-saline areas.