skip to main content


Search for: All records

Creators/Authors contains: "Peng Wang"

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. The constellation of Earth-observing satellites continuously collects measurements of scattered radiance, which must be transformed into geophysical parameters in order to answer fundamental scientific questions about the Earth. Retrieval of these parameters requires highly flexible, accurate, and fast forward and inverse radiative transfer models. Existing forward models used by the remote sensing community are typically accurate and fast, but sacrifice flexibility by assuming the atmosphere or ocean is composed of plane-parallel layers. Monte Carlo forward models can handle more complex scenarios such as 3D spatial heterogeneity, but are relatively slower. We propose looking to the computer graphics community for inspiration to improve the statistical efficiency of Monte Carlo forward models and explore new approaches to inverse models for remote sensing. In Part 2 of this work, we demonstrate that Monte Carlo forward models in computer graphics are capable of sufficient accuracy for remote sensing by extending Mitsuba 3, a forward and inverse modeling framework recently developed in the computer graphics community, to simulate simple atmosphere-ocean systems and show that our framework is capable of achieving error on par with codes currently used by the remote sensing community on benchmark results. 
    more » « less
    Free, publicly-accessible full text available March 1, 2025
  2. The constellation of Earth-observing satellites continuously collects measurements of scattered radiance, which must be transformed into geophysical parameters in order to answer fundamental scientific questions about the Earth. Retrieval of these parameters requires highly flexible, accurate, and fast forward and inverse radiative transfer models. Existing forward models used by the remote sensing community are typically accurate and fast, but sacrifice flexibility by assuming the atmosphere or ocean is composed of plane-parallel layers. Monte Carlo forward models can handle more complex scenarios such as 3D spatial heterogeneity, but are relatively slower. We propose looking to the computer graphics community for inspiration to improve the statistical efficiency of Monte Carlo forward models and explore new approaches to inverse models for remote sensing. In Part 1 of this work, we examine the evolution of radiative transfer models in computer graphics and highlight recent advancements that have the potential to push forward models in remote sensing beyond their current periphery of realism. 
    more » « less
  3. We investigated the optimal number of independent parameters required to accurately represent spectral remote sensing reflectances (Rrs) by performing principal component analysis on quality controlledin situand syntheticRrsdata. We found that retrieval algorithms should be able to retrieve no more than four free parameters fromRrsspectra for most ocean waters. In addition, we evaluated the performance of five different bio-optical models with different numbers of free parameters for the direct inversion of in-water inherent optical properties (IOPs) fromin situand syntheticRrsdata. The multi-parameter models showed similar performances regardless of the number of parameters. Considering the computational cost associated with larger parameter spaces, we recommend bio-optical models with three free parameters for the use of IOP or joint retrieval algorithms.

     
    more » « less
  4. When training overparameterized deep networks for classification tasks, it has been widely observed that the learned features exhibit a so-called “neural collapse” phenomenon. More specifically, for the output features of the penultimate layer, for each class the within-class features converge to their means, and the means of different classes exhibit a certain tight frame structure, which is also aligned with the last layer’s classifier. As feature normalization in the last layer becomes a common practice in modern representation learning, in this work we theoretically justify the neural collapse phenomenon under normalized features. Based on an un-constrained feature model, we simplify the empirical loss function in a multi-class classification task into a nonconvex optimization problem over the Riemannian manifold by constraining all features and classifiers over the sphere. In this context, we analyze the nonconvex landscape of the Riemannian optimization problem over the product of spheres, showing a benign global landscape in the sense that the only global minimizers are the neural collapse solutions while all other critical points are strict saddle points with negative curvature. Experimental results on practical deep networks corroborate our theory and demonstrate that better representations can be learned faster via feature normalization. Code for our experiments can be found at https://github.com/cjyaras/normalized-neural-collapse. 
    more » « less
  5. Abstract. Ocean color remote sensing is a challenging task over coastal watersdue to the complex optical properties of aerosols and hydrosols. Inorder to conduct accurate atmospheric correction, we previously implementeda joint retrieval algorithm, hereafter referred to as the Multi-Angular Polarimetric Ocean coLor (MAPOL) algorithm,to obtain the aerosol and water-leavingsignal simultaneously.The MAPOL algorithm has been validated with syntheticdata generated by a vector radiative transfer model, and good retrievalperformance has been demonstrated in terms of both aerosol and oceanwater optical properties (Gao et al., 2018).In this work we applied the algorithm to airborne polarimetricmeasurements from the Research Scanning Polarimeter (RSP) over bothopen and coastal ocean waters acquired in twofield campaigns: the Ship-Aircraft Bio-Optical Research (SABOR) in2014 and the North Atlantic Aerosols and Marine Ecosystems Study(NAAMES) in 2015 and 2016. Two different yet related bio-opticalmodels are designed for ocean water properties. One model aligns withtraditional open ocean water bio-optical models that parameterize theocean optical properties in terms of the concentration of chlorophyll a. The other is a generalized bio-optical model for coastal watersthat includes seven free parameters to describe the absorption andscattering by phytoplankton, colored dissolved organic matter, andnonalgal particles. The retrieval errors of both aerosol opticaldepth and the water-leaving radiance are evaluated. Through thecomparisons with ocean color data products from both in situmeasurements and the Moderate Resolution Imaging Spectroradiometer(MODIS), and the aerosol product from both the High SpectralResolution Lidar (HSRL) and the Aerosol Robotic Network (AERONET), the MAPOL algorithm demonstrates both flexibility and accuracy in retrievingaerosol and water-leaving radiance properties under various aerosoland ocean water conditions.

     
    more » « less