An Object-based Fusion Of Quickbird Data And RADARSATt SAR Data For Classification Analysis

The speckle noise is an inherent defect in SAR system, which hindered the subsequent application of SAR images so speckle noise suppression is an important step in SAR image interpretation before analysis. SAR image carries information quite different from optical image. It is very useful for the interpretation of remote sensing image and identify the forest types by using the fusion of optical remote sensing image and SAR image and complement each other. In this paper, two aspects of SAR image filtering and multi-source information fusion were chosen as the main topic.Traditional speckle noise filtering algorithm can remove the noise while can’t maintain the edge detail and to deal with the shortage, we propose an adaptive despeckle filter that can preserve edges based on noise identification oriented to SAR images. This paper first describes the mechanism of SAR image speckle noise,model and statistical properties and then according to the specific purpose and the requirements of image to analysis common speckle filters, LEE and enhanced LEE filtering, FROST and enhanced FROST filtering, GAMMA filtering, KUAN filtering, and LOCAL SIGMA filtering were applied to Jiang Le state forest farm in RADARSAT-2 images. Finally, the result were analyzed based on the evaluation index, the evaluation index include Equivalent number of looks(ENL),Edge preserving index(EPI).The results show that the enhanced LEE adaptive filter has the best performance, which can remove the speckle noise well and maintain the image of high-frequency information and details at the same time. Systematic comparative and analysis of each filtering algorithm results, we can provide a theoretical basis of selecting the SAR image classification of forest types subsequent application filtering method.Based on the Quickbird data and Radarsat-2 full polarization data of San Ming City, Fujian Province, the object-oriented method was adopted to identify the land cover types from the fusion of Quickbird panchromatic and multi-spectral image, SAR images and the fusion of Quickbird image and SAR image which acquired by using the Gram-Schmidt method. Classification factors including spectral, texture and geometric features were used to establish a class hierarchy, and the classification results were compared. The results showed that the knowledge-based and object-based methods was effective in identification and classification a high spatial resolution images, vegetation types were effectively identified. Among them, the accuracy of multi-source remote sensing data is up to 0.903. And compared to the single source of optical remote sensing and SAR remote sensing,the classification accuracy have some improvements.