Multiple Signal Classification Algorithm Combined with Volume Reflectivity Models to Improve Accuracy of the Estimated Vegetation Height in Synthetic Aperture Radar Tomography

Abstract

The aim of this paper lies in improving the accuracy of multiple signal classification (MUSIC) inversion in Synthetic Aperture Radar Tomography (TomoSAR), while using scattering statistical models. We propose new combination algorithms between MUSIC inversion and scattering statistical models. We exploited three volume scattering models, the uniform model, the exponential model, and the Gaussian model. For each probability model, the analytical expression of the corresponding inversion was computed. In order to verify the proposed method, we exploited the dataset of the BioSAR-2 project. The data was acquired in a boreal forest located in north Sweden. The attained results for the suggested new approaches were analyzed quantitatively by computing the detection rate corresponding to the area under study according to the relative error measured for the vegetation height. Qualitatively, by evaluating for each algorithm, the generated digital surface model (DSM), the relative error, and the histograms of selected zone with strong forest densities. It was shown that combining MUSIC inversion and the uniform probability model, we achieved the highest detection rate of 60.7% for a 0.3 relative error. For the exponential distribution, we obtained a detection rate of 60.2%, and the detection rate for the Gaussian distribution was 54%. For the standard MUSIC, it achieved a weak detection rate of 25.5% for a 0.3 relative error, and for the standard CAPON, it achieved a detection rate of 38.6% for a 0.3 relative error. These results indicate that the proposed approach increases the achievement of the MUSIC inversion by 35.2%, and outperforms the standard CAPON by 22.1%. This shows the importance of using probability models in MUSIC inversion for a better estimation of vegetation height in SAR tomography.