Browsing by Author "Liu, Weidi"

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Compressed Sensing Spectroscopy and Dynamic-Rate Neural Network Classification
    (2021-10-07) Liu, Weidi; Kelly, Kevin F.
    Compressive sensing (CS) is an approach for efficient signal acquisition based on the compressibility of the data and allowing the reconstruction signals below the Shannon-Nyquist sampling rate. Besides imaging, compressive sensing can be exploited in spectrometer design. Here we have constructed and compared a broadband, high resolution Echelle spectrometer against a potentially higher resolution, narrower band Sagnac Fourier spectrometer that also has larger signal intensity. The goal of each is a field-deployable Raman gas isotope system. The initial demonstration will have hardware resolution and algorithms specific to oxygen isotope identification. The end goal is also detection and recognition directly on the compressive measurements without needing to first reconstruct the entire spectrum. Towards that end, we have also combined the mathematics of manifold secants in multiple network structures to construct a detection and classification algorithm useful for spectra, as well as images. Combining this with dynamic-rate training scheme, we reach higher classification accuracy and need only a few measurement rates for training for the chosen dataset with a single neural network, rather than training a unique neural network for each rate.
  • Thumbnail Image
    Item
    Developing Compressive Sensing in Electron Microscopy and Gas Phase Spectroscopy
    (2023-04-21) Liu, Weidi; Kelly, Kevin F.
    Compressive sensing is a technology that allows the original signal to be reconstructed with far fewer multiplexed measurements than the Shannon-Nyquist sampling rate. Compared to conventional methods, it can provide a comparable quality result with a much lower sampling time, while offering the advantage of higher signal-to-noise ratio. The first part of the work focuses on research using manifold secant patterns for classification in the field of Cryogenic electron microscopy (cryo-EM). In certain scenarios, the main focus may not be on reconstructing the original signal, but rather on solving an inference problem in isolation, which classification is an example and offers certain advantage over current methods. Next, the building of a high-resolution compressive Raman spectrometer for gas analysis will be described. A novel cavity-enhanced gas Raman signal generation system was built and developed in a Sagnac spectrometer configuration. Incorporating the idea of compressive sensing into the system further enhances the signal-to-noise ratio and reduces the acquisition time as well as allowing for compactness and portability for use in remote environments. Lastly, improvements and the future potential of this work will be discussed.
  • Thumbnail Image
    Item
    General neural network approach to compressive feature extraction
    (Optical Society of America, 2021) Giljum, Anthony; Giljum, Anthony; Liu, Weidi; Liu, Weidi; Li, Le; Weber, Reed; Kelly, Kevin F.
    Computer vision with a single-pixel camera is currently limited by a trade-off between reconstruction capability and image classification accuracy. If random projections are used to sample the scene, then reconstruction is possible but classification accuracy suffers, especially in cases with significant background signal. If data-driven projections are used, then classification accuracy improves and the effect of the background is diminished, but image recovery is not possible. Here, we employ a shallow neural network to nonlinearly convert from measurements acquired with random patterns to measurements acquired with data-driven patterns. The results demonstrate that this improves classification accuracy while still allowing for full reconstruction.