Binary Classification of an Unknown Object through Atmospheric Turbulence Using a Polarimetric Blind-Deconvolution Algorithm Augmented with Adaptive Degree of Linear Polarization Priors THESIS

In this research, an improved binary material-classification algorithm is developed to discriminate between metals and dielectrics using passive polarimetric imagery degraded by atmospheric turbulence. The technique implements a modified version of an existing polarimetric blind-deconvolution algorithm in order to remove atmospheric distortion and correctly classify the unknown object. The classification decision is based on degree of linear polarization (DoLP) estimates provided by the blind-deconvolution algorithm augmented with two DoLP priors—one statistically modeling the polarization behavior of metals and the other statistically modeling the polarization behavior of dielectrics. The proposed algorithm significantly improves upon a similar published polarimetric classification method by adaptively updating the DoLP priors as more information becomes available about the scene. Three approaches for implementing the adaptive DoLP priors are presented— the higher-order super-Gaussian method, the Gaussian method, and the distributionaveraging method. The higher-order super-Gaussian method fits the distribution of the in-progress DoLP estimates from the blind-deconvolution algorithm with a sum of two super-Gaussian functions. The results of the nonlinear fit are then used to form the DoLP priors. The Gaussian method fits the distribution of DoLP estimates with the sum of two Gaussian functions to compute a classification threshold value. The resulting threshold is then used to update the DoLP priors. The distributionaveraging method approximates the threshold value by finding the mean of the DoLP distribution. Using this threshold value, the DoLP priors are then formed. The proposed technique is experimentally validated by comparing classification results of a dielectric and metallic sample obtained using the new method to those obtained using the existing approach. The experimental results confirm that the new adaptive method significantly extends the range of validity of the existing polarimetric