This talk reviews approaches to utilize the single pixel camera for specific machine vision tasks directly on compressive measurements. Included are simulation and hardware results of coupling an optical modulator to small infrared focal plane arrays and visible spectrometers to perform high-resolution object recognition.
The combination of wideband spectrum sensing, optical RF signal buffering, low-latency digital processing, and fast-tuning digital RF receivers enables reliable capture of transient RF events across extremely wide bandwidths for comprehensive situational awareness in electronic warfare and other defense applications.
We introduce a sparse-aperture variant of the Linear Sampling Method, which is a qualitative inverse scattering technique for reconstructing target shape. The technique reduces imaging artifacts arising from sparse data collections by incorporating a priori knowledge of propagation-based phase-delay into the inversion.
Convolution Neural Networks (CNN) are artificial networks able to extract features from large dataset by spatial filtering. Here we propose an optical coprocessor able to perform large image filtering and convolutions based on a two stage 4F system and digital micromirror arrays, outperforming current architectures.
We demonstrate a method of compressive single-pixel imaging that allows for spatial foveation anywhere in the image, determined after data acquisition, using the Sum-To-One Transform. We also show a novel method of generating fast L2 previews from Sum-To-One pattern measurements.