مقایسۀ قابلیت تصاویر ماهواره‌های سنتینل 2 و لندست 8 در تهیۀ نقشۀ کاربری اراضی و پوشش زمین با استفاده از روش‌های طبقه‌بندی پیکسل‌پایه و شی‌پایه

داشتن اطلاعات دقیق و بهنگام از وضعیت کاربری اراضی و پوشش زمین، یک نکتۀ کلیدی برای حفاظت و مدیریت پایدار منابع طبیعی است. به‌رغم اینکه ماهواره سنتینل 2 اخیراً به فضا پرتاب شده تاکنون پژوهش‌های کافی در زمینۀ به‌کارگیری تصاویر این ماهواره صورت نگرفته است. به همین دلیل در تحقیق حاضر سعی شد قابلیت تصاویر این ماهواره در تهیۀ نقشۀ کاربری اراضی و پوشش زمین منطقۀ بسطلام، استان لرستان با استفاده از روش‌های طبقه‌بندی پیکسل‌پایه و شی‌پایه بررسی و با تصاویر ماهوارۀ لندست 8 مقایسه شود. به‌منظور انجام این پژوهش در طبقات رودخانه، کشاورزی، جنگل، باغ، مسکونی، بایر و مرتع به‌ترتیب 90، 280، 680، 59، 180، 85 و 70 نمون/ آموزشی توسط بازدید میدانی، استفاده از نقشه‌های موجود، ایجاد تصاویر رنگی کاذب و نرم‌افزار گوگل ارث جمع‌آوری شد. تصاویر پس از انجام پیش‌پردازش‌های لازم، با استفاده از الگوریتم‌های حداکثر احتمال، حداقل فاصله و فاصلۀ ماهالانوبیس روش‌ پیکسل‌پایه و الگوریتم نزدیک‌ترین همسایه روش شی‌پایه طبقه‌بندی شد. میزان صحت کلی و ضریب کاپا برای طبقه‌بندی تصاویر ماهوارۀ سنتینل 2 با استفاده از روش‌های حداکثر احتمال، فاصلۀ ماهالانوبیس و حداقل فاصله به‌ترتیب 85.2% و 0.77، 88.77% و 0.81، 71.1% و 0.56 و برای تصاویر ماهوارۀ لندست 8 به‌ترتیب 84.10 % و 0.76، 82.83 % و 0.73، 63.50 % و 0.48 محاسبه شد. همچنین نتایج طبقه‌بندی شی‌پایه به روش نزدیک‌ترین‌همسایه نشان داد که تصاویر سنتینل 2 با صحت کل 89.70% و ضریب کاپا 0.83 عملکرد بهتری نسبت به تصاویر لندست 8 با صحت کل 88.90% و ضریب کاپا 0.80، داشته است. در نهایت با توجه به نتایج به‌دست‌آمده مشخص شد که استفاده از تصاویر ماهوارۀ سنتینل 2 و روش طبقه‌بندی شی‌پایه باعث بهبود نتایج طبقه‌بندی پوشش زمین و کاربری اراضی شده، اما میزان صحت نتایج استفاده از تصاویر ماهوارۀ لندست 8 و روش پیکسل‌پایه نیز قابل قبول است.

Comparing the Capability of Sentinel 2 and Landsat 8 Satellite Imagery in Land Use and Land Cover Mapping Using Pixel-based and Object-based Classification Methods

Introduction: Having accurate and uptodate information on the status of land use and land cover change is a key point to protecting natural resources, sustainable agriculture management and urban development. Preparing the land cover and land use maps with traditional methods is usually time and cost consuming. Nowadays satellite imagery provides the possibility to prepare these maps in less time and with greater accuracy. In order to prepare land use and land cover maps using satellite imagery, different classification methods are used. In a general division, these methods can be divided into two groups of pixelbased and objectbased classification methods. Despite the fact that the Sentinel 2 satellite has recently been launched and its images have been freely available to users, there has not been enough research on the use of this satellite imagery in land use and land cover mapping. Therefore, in present research, we tried to investigate the capability of Sentinel 2 satellite imagery in land use and land cover mapping of Bastam basin in Lorestan province, using pixelbased and objectbased classification methods and compare it with Landsat 8 satellite imagery.Material and methods: In order to conduct this research, training samples were collected by field survey, using existing maps, creation of false color images and Google Earth software in six classes including garden, bare land, forest, agriculture, residential and river. The 90, 280, 680, 59, 180, 85 and 70 training samples were collected in the river, agriculture, forest, garden, residential, bare land and pasture classes, respectively. The 70% of collected samples were used as training samples and 30% of the rest were used as test samples. After performing the necessary preprocessing, images classified using the Maximum likelihood, Minimum distance, Mahalanobis distance algorithms of pixelbased method and the Nearest neighbor algorithm of objectbased method. The objectbased classification method consists of two main stages of segmentation and classification. In this research, Multi Resolution Segmentation algorithm was used for segmentation. Classification was done by the Nearest Neighbor method in eCognition software. Accuracy assessment of classification results was performed using test samples and overall accuracy, Kappa coefficient, user accuracy and producer accuracy indices.Results: The overall accuracy and Kappa coefficient for Sentinel 2 satellite images were classified using Maximum likelihood, Mahalanobis distance and Minimum distance were calculated (85.2% and 0.77), (88.77% and 81.8) and (71.1% and 0.56) respectively. For Landsat 8 satellite images were classified using Maximum likelihood, Mahalanobis distance and Minimum distance the overall accuracy and kappa coefficient were observed (84.10% and 76%), (82.83% and 0.73) and (63.50% and 48.3%) respectively, which indicates a relatively higher potential Sentinel 2 image compared to Landsat 8 images in the pixel classification of land use and land cover. In addition, the results of the objectbased classification by the Nearest neighbor method showed that the Sentinel 2 images with overall accuracy 89.70% and kappa coefficient 0.83 had better performance than Landsat 8 images with overall accuracy 88% and Kappa coefficient 0.81.Discussion and conclusion: According to the results, it can be concluded that of Sentinel 2 satellite images have a better relative performance than Landsat 8 satellite imagery in the preparation of land use and land cover maps, this seems to be due to the higher spatial resolution of the Sentinel 2 satellite (10meter pixel in near infrared and visible bands) compared to Landsat 8 (30meter pixel in near infrared and visible bands). The highest level of user and producer accuracy were observed in the forest land cover, it seems that the reason for this is the vast forest area in the study area and as a result of increasing the number of training samples in this land cover. Comparison of the performance of the pixelbased and objectbased classification methods showed that the use of the objectbased classification method improves the results of land use and land cover classification. Finally, it is important to note that, despite the better performance of Sentinel 2 satellite imagery and object based classification method than Landsat 8 satellite images and pixelbased classification method, but the results of Landsat 8 satellite imagery, and pixelbased method was acceptable as well.