Angle-independent spectropolarimetric target classification using machine learning

A method for characterizing unknown targets using a hyper-spectral polarimetric light detection and ranging (LiDAR) system is presented. Light reflected from manmade objects tends to be polarized differently than light reflected from objects in the natural world. As such, polarization measurements can be used in remote sensing applications to differentiate artificial and natural objects. Here, a k-nearest neighbors (KNN) algorithm is provided for classifying materials using hyper-spectral LiDAR with incident angles ranging from -10^∘ to 60°, without assuming any further information about the orientation of the target relative to the LiDAR system, achieving a balanced accuracy of 86.9%.