ارزیابی تطبیقی تکنیک های پردازش پیکسل پایه و شیءگرا در طبقه بندی تصاویر ماهواره ای aster برای استخراج نقشه های اراضی کشاورزی و باغی در حاشیه شرقی دریاچه ارومیه

تحقیق حاضر نمونه‌ای از کاربرد تکنولوژی سنجش از دور در مدیریت منابع کشاورزی است. در این تحقیق با پردازش رقومی تصاویر ماهواره‌ای aster خرداد ماه سال 2016 نقشه‌های کاربری اراضی حاشیه شرقی دریاچه ارومیه استخراج شده است. در این ارتباط در مرحله پیش‌پردازش، تصحیحات هندسی شامل زمین مرجع کردن، تصحیحات ارتفاعی و تصحیحات اتمسفری برروی تصاویر اعمال شد. در مرحله پردازش، پس از اعمال توابع آشکارسازی، متناسب با اهداف پژوهش طبقه‌بندی براساس الگوریتم‌های شیءگرا و پیکسل پایه برروی تصاویر انجام شد. برای این منظور از الگوریتم‌های حداکثر احتمال، متوازی السطوح و حداقل فاصله از میانگین تصاویر طبقه‌بندی استفاده شد. سپس، پردازش شیءگرای تصاویر ماهواره‌ای برروی تصاویر اعمال گردید. در این راستا، در ابتدا فرایند سگمنت‌سازی برروی تصاویر انجام شد و تصاویر متناسب با معیارهای همگنی، ضریب شکل و فشردگی مورد سگمنت‌سازی قرار گرفتند. طبقه‌بندی از نوع شیء‌گرا با استفاده از الگوریتم‌های طیفی و مکانی و روش نزدیکترین همسایگی در محیط نرم افزارecognition طی مراحل مختلف پیاده شد.به منظور ارزیابی و مقایسه نتایج، ضرایب دقت کلی و کاپای طبقه‌بندی برای هر کدام از الگوریتم‌ها استخراج و مشخص شد که در میان روش‌های پیکسل پایه، الگوریتم طبقه‌بندی حداکثر احتمال با ضریب کاپای 0.86و دقت کلی 88.67 درصد در مقایسه با سایر روش‌ها، از دقت بالاتری برخوردار است. اما خود این الگوریتم نیز در مقایسه با روش شیءگرا از دقت کمتری برخوردار است، چرا که ضریب کاپای طبقه‌بندی حاصله معادل 0.93 و دقت کلی نیز معادل 94.20 درصد برآورد گردید.

A comparative evaluation of PixelBased and ObjectOriented processing techniques, used for the classification of Aster Satellite imageries and extracting agricultural and orchard maps in the Eastern Margin of Urmia Lake

IntroductionNowadays, detailed land cover and land use information is considered to be an important research topic in geosciences, environmental changes and natural resources.  In this regard, monitoring agricultural landuse provides essential information for land use planners and decision makers. Multiple methods are used for monitoring agricultural land use from remotely sensed images. Objectoriented techniques used for processing satellite images makes high accuracy recognition of various land use patterns possible. Compared to traditional pixelbased approaches, these techniques reach a higher accuracy in extracting landuse information from satellite imageries using geometric information and features of different phenomena. In fact, object oriented methods relay on processing unit or image objects made by the integration of homogenized pixels in the segmentation process. Once segments are formed, various indices and spatial information like texture, pattern, form, content, and etc. are applied on processing units. In this way, identifying appropriate land use classes based on geometric features becomes possible. As compared to traditional pixelbased methods, these methods are more flexible and thus can apply a combination of spectral and spatial information. The main goal of the present article is to develop landuse maps and to evaluate agricultural activities in eastern basin of Urmia Lake using objectoriented processing techniques. Study area and martial The study area was chosen based on a mixture of agricultural land use, human settlements, and salt marshes in Urmia lake eastern margins. The main goal of the present study was to produce accurate maps of agricultural systems in Urmia lake eastern margins. Thus, various agricultural land uses were extracted from satellite imageries with an emphasis on orchard land use classes. Training data were collected through field operation using GPS. Moreover, 1:25000 scale topographic maps were used for geometric correction and rectifying of the satellite images. Methods and techniques Various agricultural and orchard land use were extracted from Aster satellite imageries received in 2016. In preprocessing stage, geometric correction including georeferencing, orthorectification and atmospheric corrections were performed on imageries. In processing stage, detection functions were applied and images were then classified according to the research goals based on pixelbased and objectoriented algorithms. In this regard, maximum likelihood, parallelepiped, and minimum distance algorithms were used to classify images. Segmentation process was performed based on homogeneity, shape and compactness parameters. Accordingly, the geometric and spectral algorithms were used for modeling each class in objectoriented environment and objectoriented classification was applied based on nearest neighbor algorithm. ResultsUsing object oriented and pixel based processing techniques, four land use maps were extracted. In order to evaluate and compare final results, overall accuracy and Kappa coefficients were extracted for each algorithm. Results indicate that among pixelbased classification algorithms, maximum likelihood algorithm with overall accuracy of 87.67 percent and kappa coefficient of 0.86 is more accurate than other methods. However, with a Kappa coefficient of about 0.93 and overall accuracy of 94.20 percent, this algorithm has a lower accuracy level as compared to objectoriented methods. Discussion and conclusion Results indicate that compared to pixel based techniques, object oriented processing techniques possesses a higher potentiality for extracting agricultural land use. The main advantage of object oriented methods is that they employ a combination of spatial information, spectral information and integrate them with GIS and remote sensing datasets. Moreover, using texture and shape algorithms in object based classification leads to improved accuracy of land use maps. Besides, it is possible to improve the accuracy of results using effective techniques in object oriented classification. According to research findings, object oriented techniques provide an effective method for classification of satellite imageries and extraction of land use maps. It is possible to use these techniques in landscape planning, natural resources, regional landuse and landcover changes, sustainability of land cover, and etc.