Dynamic Antenna Array Design for Scene Classification Through Fourier-Domain Filtering

We present a new approach to the classification of scenes in the microwave and millimeter-wave bands that leverages a novel dynamic antenna array concept to capture distinct features in the spatial frequency information of the scene. The spatial frequency information of a scene is obtained through its Fourier transform, and by sampling a subset of this information, key features can be extracted and used for image classification. We demonstrate that a dynamic antenna array can synthesize spatial frequency filters, and that scene classification can be accomplished using the filtered signals without full image reconstruction. We develop a new dynamic antenna array concept using only two antennas to generate a ring-shaped spatial frequency filter and explore the use of this concept for the classification of ground scenes. Natural ground scenes tend to have smoother spatial frequency signals, while, in contrast, features such as buildings and roadways result in sharp broadband spatial frequency responses. Using this design, we demonstrate the ability to classify between two classes of ground scenes: those with man-made structures (buildings, roads, etc.) and those without (natural scenes). We demonstrate the ability of the spatial filters synthesized by the proposed dynamic antenna array to achieve a classification accuracy of 0.971 with an empirical true positive rate of 0.982. The method is broadly applicable to microwave and millimeter-wave sensing at any range.