Spatial Frequency Filter Design for Interferometric Image Classification Without Image Reconstruction

We investigate the use of spatial frequency filtering to detect specific features and classify images without reconstructing a full image. Based on interferometric Fourier-domain imaging, filtering the spatial frequency information amounts to a data reduction at the input to the system, leading to lower computational complexity, less hardware requirements, and the ability to classify images without the need for full image reconstruction. The proposed application is the detection of man-made structures from interferometric microwave imagery of the ground. In the spatial frequency domain, man-made structures such as buildings and roads display discrete, high spatial-frequency signals, while natural scenes have a smoother spatial frequency profile. We present ring-shaped spatial frequency designs that can detect these features without full image reconstruction. Furthermore, the filters can potentially be implemented with a small set of antennas, leading to low-cost, fast classification imaging.