Classification of levees using polarimetric Synthetic Aperture Radar (SAR) imagery

The recent catastrophe caused by hurricane Katrina emphasizes the importance of examination of levees to improve the condition of those that are prone to failure during floods. On-site inspection of levees is costly and time-consuming, so there is a need to develop efficient techniques based on remote sensing technologies to identify levees that are more vulnerable to failure under flood loading. This research uses NASA JPL's Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) backscatter data for classification and analysis of earthen levees. The overall purpose of this research is to detect the problem areas along the levee such as through-seepage, sand boils and slough slides. This paper focuses on detection of slough slides. Since the UAVSAR is a quad-polarized L-band (λ = 25 cm) radar, the radar signals penetrate into the soil which aids in detecting soil property variations in the top layer. The research methodology comprises three steps: initially the SAR image is classified into three scattering components using the Freeman-Durden decomposition algorithm; then unsupervised classification is performed based on the polarimetric decomposition parameters: entropy (H) and alpha (α); and finally reclassified using the Wishart classifier. A 3×3 coherency matrix is calculated for each pixel of the radar's compressed Stokes matrix multi-look backscatter data and is used to retrieve these parameters. Different scattering mechanisms like surface scattering, dihedral scattering and volume scattering are observed to distinguish different targets along the levee. The experimental results show that the Wishart classifier can be used to detect slough slides on levees.