Identification and separation of land covers by optical and RADAR image fusion

Mostafa Kabolizadeh ¹ ( RS and GIS department, Earth science faculty, shahiud Chamran university of Ahvaz )
Sajad Zareie ² ( Department of Remote Sensing and GIS, Faculty of Earth Sciences, Shahid Chamran University of Ahvaz, Ahvaz, Iran )
Rahman Khanafereh ³ ( Department of Remote Sensing and GIS, Faculty of Earth Sciences, Shahid Chamran University of Ahvaz, Ahvaz, Iran )

Submited date : 2022-12-24 Accepted date : 2023-01-28

Keywords: support vector machine (SVM) method, texture features, Classification, Feature Selection, land covers in urban environments,

Abstract :

Classification and separation of land cover is one of the most important applications of remote sensing. To perform classification, multispectral satellite data are an efficient tool, but unfortunately, they are not available in some conditions, such as cloudy weather. Also, most remote sensing data classification algorithms operate based on the characteristics and spectral information of pixels, which causes the useful spatial information that can be extracted from the images to be ignored, including; The texture of the pictures. The simultaneous use of texture and spectral information is a topic that has been less discussed. Therefore, considering this idea, two methods were used to select the optimal features for preparing the land cover map. The first method is the normalized reflection of complications according to the extracted features and the second method is applying the Optimum Index Factor on the extracted textural and spectral features. For this purpose, the classification process using the Support Vector Machine method, on the Sentinel-1 radar image and the Sentinel-2 multispectral image, the optimal features selected by the two methods and the combination of image bands with the optimal features selected by It was done by two methods and finally by combining the best combination of radar and optical bands. According to the obtained results, the classification using spectral features is more accurate than the classification using texture features. By combining the optical and radar features and obtaining values of 97.07% for the overall accuracy and 0.96% for the Kappa coefficient, the classification accuracy was improved to a great extent. This research showed that by choosing optimal features and combining spectral and radar data, different features of each data can be used and better results can be achieved.