نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشگاه یزد، یزد، ایران
2 استادیار، گروه جغرافیا بخش برنامه ریزی محیطی، دانشگاه یزد، یزد،ایران،
3 گروه جغرافیای طبیعی، دانشکده علوم جغرافیایی و برنامه ریزی، دانشگاه اصفهان، اصفهان، ایران
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
Introduction: Nowadays, satellite imagery enables the classification of land cover, which is one of the key approaches in analyzing environmental changes, managing natural resources, assessing land use dynamics, and monitoring natural hazards. These methods assist urban planners, land resource managers, and decision-makers in obtaining accurate information about the current state and trends of environmental changes. Among these methods, unsupervised classification techniques are widely used due to their independence from labeled training data and their suitability for analyzing large and complex datasets. Objective: The main objective of this research is to design and implement a fast, organized, and accurate unsupervised classification method for Landsat 8 satellite imagery that can be executed in the shortest possible time. The proposed method is based on various spectral indices and combines the K-means algorithm with a simple artificial neural network (ANN), designed in such a way that it does not require any training data. Methodology: In this study, the Landsat 8 satellite image containing 10 main bands (7 multispectral, 2 thermal, and 1 panchromatic) was preprocessed. Several indices were then extracted, including the Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), Normalized Difference Water Index (NDWI), Soil Adjusted Vegetation Index (SAVI), and Dry Bare Soil Index (DBI). Additional features such as brightness, texture, and edge information were also extracted to improveclassseparability.After data normalization, the K-means algorithm was applied for initial clustering, and the results were used as weak labels for the neural network. The designed neural network contained a single hidden layer and received the extracted features as input to refine the clusters and produce the final classification map. The thermal bands were used to distinguish warm and dry soil areas from shaded and cooler regions, while the panchromatic band was utilized to accurately detect urban and suburban road networks. Results: The final classification results demonstrated that the proposed method achieved an overall accuracy of 80.36%, which is considered acceptable. The algorithm successfully distinguished rooftops, barren mountains, bare soils, unpaved areas, and both urban and interurban roads with clear boundaries. The amount of noise in the final unsupervised classification output was significantly reduced, and the spatial consistency of the classes was preserved. Qualitative evaluation in the Sabzevar region showed that the improved unsupervised classification method could accurately identify vegetation, built-up areas, and barren lands with high spatial detail. Furthermore, it effectively separated urban streets from interurban roads. Conclusion: In this research, by utilizing spectral indices such as NDVI, NDBI, NDWI, thermal bands, and an artificial band, along with a simple artificial neural network, an efficient unsupervised classification algorithm was developed. The main goal was to achieve accurate land cover separation in urban and mountainous areas. Despite not using any training data or pre-existing algorithms, the improved unsupervised classification method successfully delineated similar classes such as urban streets, interurban roads, barren mountains, and rooftops with high accuracy. Although the overall accuracy (80.36%) and Kappa coefficient (0.72) were slightly lower than those of traditional algorithms such as SVM and Decision Tree, the spatial quality and class separability of the proposed method were considerably superior. The output maps produced by the improved algorithm are more realistic both visually and analytically.
کلیدواژهها [English]
)