Schematic overview. Information of interest χ, “3” is physically encoded in a scene reflectivity of a corresponding handwritten metal digit. The scene is probed with M measures. For each measurement, a TX DMA radiates a coherent field to the scene, and the reflection is coherently captured by an RX DMA, producing a single complex-valued scalar mI which is corrupted by noise nI. The coherently radiated and captured fields of the TX and RX DMAs in the ith measurement are determined by the configuration of the meta-atoms of the DMAs, CITX etcIX-ray, respectively. The wave energy injected into the TX DMA is always the same. The measured data from the M measurements are injected into a fully connected digital processing neural network in order to output an estimate χ˜ of the information of interest sought. The overall procedure is therefore parameterized by physical weights (meta-atom configurations) and numerical weights (numerical ANN weights). The inset shows the DMA hardware under consideration: a 2D parallel plate waveguide with 16 programmable 1-bit meta-atoms patterned into the surface facing the stage. Depending on the bias voltage applied across the PIN diode, a given meta-atom is either resonant (blue) or non-resonant (red), as shown for an example configuration on the left. The corresponding dipole moments of the meta-atoms are shown on the right as phasors. Credit: Intelligent Computing (2022). DOI: 10.34133/2022/9825738
Sensing systems are becoming more prevalent in many areas of our lives, such as ambient healthcare, autonomous vehicles, and contactless human-computer interaction. However, these systems often lack intelligence: they tend to gather all available information, even if it is not relevant. This can lead not only to privacy breaches, but also to a waste of time, energy and computer resources when processing data.
To solve this problem, researchers from the French CNRS have proposed a concept of intelligent electromagnetic sensing, which uses machine learning techniques to generate learned lighting patterns in order to pre-select relevant details during the measurement process. A programmable metasurface is configured to generate the learned patterns, performing high-precision sensing (eg, posture recognition) with a remarkably reduced number of measurements.
But measurement processes in realistic applications are inevitably subject to a variety of noises. Noise basically accompanies any measurement. The signal-to-noise ratio can be particularly low in indoor environments where radiated electromagnetic signals must be kept low.
Therefore, Chenqi Qian and Philipp del Hougne extended their previous research and presented an intelligent programmable computer meta-imager that not only adapts its lighting pattern to a specific information-mining task like object recognition, but also adapts to different types and levels of noise. They published a guest research paper on their results in Intelligent Computing on December 2, 2022.
“We hypothesize that the optimal coherent lighting patterns to be used by an intelligent programmable meta-imager to efficiently extract task-specific information from a scene will be deeply dependent on the type and level of noise,” said the authors. researchers, pointing out that noise can profoundly impact the optimal configurations of meta-imagers because, in addition to latency constraints that limit the number of measurements allowed, noise also limits the amount of information that can be extracted from the scene.
“In this paper, we systematically explore the impact of the combination of latency and noise constraints on multi-tap programmable smart meta-imagers,” the researchers said. To test their hypothesis, the researchers considered a prototypical object recognition problem, for which they proposed a microwave-programmable meta-imager system. Such systems could be deployed in indoor surveillance, earth observation, etc.
In their system under consideration, a microwave dynamic metasurface (DMA) antenna radiated a sequence of coherent wavefronts to the scene using a single transmitter, and a second DMA coherently captured the reflected waves at the using a single detector. A differentiable end-to-end information flow pipeline was formulated, which included the programmable physical measurement process, including noise, as well as the subsequent digital processing layers.
The essential elements of this pipeline are the same for all wave-based information extraction problems, including imaging, detection, localization, and object recognition. “The only significant difference is in the task-specific cost function that needs to be optimized for good performance,” they explained.
The same approach that the authors applied to object recognition can therefore also be used in parameter estimation problems such as localization. “This pipeline allows us to co-design the programmable physical weights (DMA configurations that determine consistent scene lights) and the trainable digital weights.”
It is this joint optimization—task-specific end-to-end joint optimization of trainable physical parameters and trainable numerical parameters—that makes the measurement process task-aware, so that it could discriminate between relevant information for the task and those which are not. air in the analog domain.
Researchers tested the performance of this programmable meta-imager that generates a sequence of task-specific and noise-specific scene illuminations and found it advantageous over conventional compressed sensing with random configurations when the information could be extracted from the scene are limited by latency. stresses and/or noise. Performance gains for a signal-independent and signal-dependent type of additive noise have both been demonstrated. The “macroscopic” characteristics of the learned lighting patterns, namely their mutual overlaps and intensities, proved to be intuitively understandable despite the “black box” nature of the approach.
Transitioning to a system that auto-adaptively detects the type and level of noise and updates its used sequence of DMA configurations accordingly without additional human intervention is straightforward, the researchers say. “We sincerely hope that our results can be transposed to information retrieval problems based on other wave phenomena (e.g., optics, acoustics, rubber bands, and quantum mechanics) and/or with d ‘other types of field-programmable measurement equipment,’ they concluded. .
Chenqi Qian et al, Noise Adaptive Smart Programmable Meta-Imager, Intelligent Computing (2022). DOI: 10.34133/2022/9825738
Powered by Intelligent Computing
Quote: Intelligent Programmable Meta-Imagers: A Timely Approach to Task- and Noise-Adapted Detection (2022, December 30) Retrieved December 30, 2022, from https://techxplore.com/news/2022-12-intelligent-programmable-meta – imagers-approach-task-specific.html
This document is subject to copyright. Except for fair use for purposes of private study or research, no part may be reproduced without written permission. The content is provided for information only.
#Smart #programmable #metaimagers #timely #approach #tasktailored #noisetailored #detection