An Overview of Neural Network Methods for Predicting Uncertainty in Atmospheric Remote Sensing

Doicu, Adrian; Doicu, Alexandru; Efremenko, Dmitry; Loyola, Diego; Trautmann, Thomas

doi:10.3390/rs13245061

Cited by 2 publications

(3 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that (i) E([∆ x − E(∆ x )] 2 ) reproduces the square root of the diagonal elements of the so-called epistemic covariance matrix of all network errors over the prediction set, and (ii) the epistemic uncertainties are large if there are large variations around the mean, e.g., if E j ([x pred − E j (x pred )] 2 ) are large. Also note that non-optimal hyper-and training parameters, as well as, a non-optimal optimization algorithm are the main sources of epistemic or model uncertainty [61]. The following conclusions can be drawn.…”

Section: A Synthetic Retrievalmentioning

confidence: 79%

“…The design and refinement of neural networks for atmospheric retrieval is a very complicated research field that requires more developments that consist of 1) application of the inverse-operator neural networks to the remaining aerosol models considered in the MODIS algorithm, i.e., non-absorbing, absorbing, and desert dust (the selection of an appropriate aerosol model is then based on a combination of spectral and geographic information); 2) training the neural networks to learn the relative evidences of different aerosol models, so that, a mean solution estimate, representing a linear combination of candidate solutions weighted by their evidences, can be computed [62]; 3) redesign of the neural networks in a Bayesian deep learning framework in order to predict input aleatoric and model uncertainties [61].…”

Section: B Retrieval From Real Datamentioning

confidence: 99%

See 1 more Smart Citation

Aerosol Parameters Retrieval From TROPOMI/S5P Using Physics-Based Neural Networks

Rao

Efremenko

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

Self Cite

View full text Add to dashboard Cite

In this paper, we present three algorithms for aerosol parameters retrieval from TROPOMI measurements in the O2 A-band. These algorithms use neural networks (i) to emulate the radiative transfer model and a Bayesian approach to solve the inverse problem, (ii) to learn the inverse model from the synthetic radiances, and (iii) to learn the inverse model from the principalcomponent transform of synthetic radiances. The training process is based on full-physics radiative transfer simulations. The accuracy and efficiency of the neural network based retrieval algorithms are analyzed with synthetic and real data.

show abstract

Section: A Synthetic Retrievalmentioning

confidence: 79%

Section: B Retrieval From Real Datamentioning

confidence: 99%

Aerosol Parameters Retrieval From TROPOMI/S5P Using Physics-Based Neural Networks

Rao

Efremenko

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, the mainstream of natural image SR methods is 2D CNN, which can lead to severe spectral distortion when applied to multi-band HSIs. Recently, some methods have emerged that combine hand-craft priors and neural networks and introduce mathematical modeling considering the properties of HSI, such as [15][16][17].…”

Section: Introductionmentioning

confidence: 99%

Rethinking 3D-CNN in Hyperspectral Image Super-Resolution

Liu

Wang

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Recently, CNN-based methods for hyperspectral image super-resolution (HSISR) have achieved outstanding performance. Due to the multi-band property of hyperspectral images, 3D convolutions are natural candidates for extracting spatial–spectral correlations. However, pure 3D CNN models are rare to see, since they are generally considered to be too complex, require large amounts of data to train, and run the risk of overfitting on relatively small-scale hyperspectral datasets. In this paper, we question this common notion and propose Full 3D U-Net (F3DUN), a full 3D CNN model combined with the U-Net architecture. By introducing skip connections, the model becomes deeper and utilizes multi-scale features. Extensive experiments show that F3DUN can achieve state-of-the-art performance on HSISR tasks, indicating the effectiveness of the full 3D CNN on HSISR tasks, thanks to the carefully designed architecture. To further explore the properties of the full 3D CNN model, we develop a 3D/2D mixed model, a popular kind of model prior, called Mixed U-Net (MUN) which shares a similar architecture with F3DUN. Through analysis on F3DUN and MUN, we find that 3D convolutions give the model a larger capacity; that is, the full 3D CNN model can obtain better results than the 3D/2D mixed model with the same number of parameters when it is sufficiently trained. Moreover, experimental results show that the full 3D CNN model could achieve competitive results with the 3D/2D mixed model on a small-scale dataset, suggesting that 3D CNN is less sensitive to data scaling than what people used to believe. Extensive experiments on two benchmark datasets, CAVE and Harvard, demonstrate that our proposed F3DUN exceeds state-of-the-art HSISR methods both quantitatively and qualitatively.

show abstract

An Overview of Neural Network Methods for Predicting Uncertainty in Atmospheric Remote Sensing

Cited by 2 publications

References 47 publications

Aerosol Parameters Retrieval From TROPOMI/S5P Using Physics-Based Neural Networks

Aerosol Parameters Retrieval From TROPOMI/S5P Using Physics-Based Neural Networks

Rethinking 3D-CNN in Hyperspectral Image Super-Resolution

Contact Info

Product

Resources

About