Exploration of Data Fusion between Polarimetric Radar and Multispectral Image Data.

摘要

Typically, analysis of remote sensing data is limited to one sensor at a time which usually contains data from the same general portion of the electromagnetic spectrum. SAR and visible near infrared data of Monterey, CA, were analyzed and fused with the goal of achieving improved land classification results. A common SAR decomposition, the Pauli decomposition was performed and inspected. The SAR Pauli decomposition and the multispectral reflectance data were fused at the pixel level, then analyzed using multispectral classification techniques. The results were compared to the multispectral classifications using the SAR decomposition results for a basis of interpreting the changes. The combined dataset resulted in little to no quantitative improvement in land cover classification capability, however inspection of the classification maps indicated an improved classification ability with the combined data. The most noticeable increases in classification accuracy occurred in spatial regions where the land features were parallel to the SAR flight line. This dependence on orientation makes this fusion process more ideal for datasets with more consistent features throughout the scene.