Covariance Based Approach Using Expectation Maximization Algorithm For Forest Height Estimation

Synthetic Aperture Radar Tomography has been an active research field for the estimation of both artificial structures and forest heights. Most spectral analysis approaches explore the covariance matrix (CM) to provide accurate reflectivity profiles. In practical situations, the statistical CM is unknown, hence several methods replace it with the sampling CM. Unfortunately, the latter does not always meet the application requirements. At our end, we propose to apply the Expectation Maximization algorithm to iteratively estimate the pseudo spectrum and then update the CM. The impact the proposed method has on height estimation over vegetated areas is carried out. Experimental results on a real dataset acquired by an airborne system show the effectiveness of the Expectation Maximization estimator in terms of ground and canopy discrimination for each polarization channel.