Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Topics - Pankaj Dey

Pages: 1 2 [3] 4 5 ... 29
33
Interesting information / Courses: Applied Remote Sensing
« on: October 11, 2019, 04:37:13 PM »
Statistical Signal Processing (60 hours) - G. Camps-Valls
Material for a master course on (statistical) signal processing. I cover the essential background for engineers and physicists interested in signal processing: Probability and random variables, discrete time random processes, spectral estimation, signal decomposition and transforms, and an introduction to information theory.
Material
Representation of spatial information (30 hours) - J. Malo
Statistical regularities in photograpic images imply that certain representations of spatial information are better than others in terms of coding efficiency. In this course we present the information theory concepts (entropy, multi-information, correlation and negentropy) for unsupervised feature extraction or dictionary learning required in image coding. Redundancy in images and sequences is reviewed, and basic techniques for compact information representation are introduced such as vector quantization, predictive coding and transform coding. Application of these concepts in images are the basis of DCT and Wavelet representations which are the core of JPEG and JPEG2000.
Material
Color vision and Colorimetry (30 hours) - J. Malo
Color is a 5-dimensional perception that is not only related to the spectrum coming from an object, but also strongly related to its spatio-temporal context. It is a powerful feature that allow humans to make reliable inferences about objects that would be nice to understand and mimic in artificial vision. In this course we derive the linear CIE tristimulus theory from its experimental color matching foundations. We derive the relations between spectrum and tristimulus vectors through the color matching functions, the chromatic coordinates, chromatic purity and luminace. Phenomenology of color discrimination and adaptation reveal the limitations of the linear description and set the foundations of color appearance models. In addition, we link the above perceptual representations of color with the conventional representation of color in computers.
Material
Texture and motion in the visual cortex (40 hours) - J. Malo
Neurons in V1 and MT cortex play a determinant role in the analysis of the shape of objects, their spatial texture and the estimation of retinal motion. In this course we describe the basic psychophysical and physiological phenomena related to low-level spatio-temporal vision: the contrast sensitivity functions, masking, adaptation and aftereffects. These facts are mediated by the context-dependent nonlinearities of the response of neurons with specific receptive fields. We analyze the geometric properties of the standard model of V1 and their consequences in image discrimination. We introduce the concept of optical flow, its properties, and how this description of motion can be estimated from the 3D wavelet sensors in V1 and the aggregated sensors in MT.
Material
Kernel methods in machine learning (30 hours) - G. Camps-Valls
Two fundamental operations in Machine Learning such as regression and classification involve drawing nonlinear boundaries or functions through a set of (labled or unlabled) training samples. These boundaries or functions at certain (test) sample can be deduced from the similarities between the test sample and the training samples. These similarities can be encoded in Kernels and the representer theorem can be used to obtain expressions for the functions at any test sample. In this course we will also review the application of the kernelization of scalar products (e.g. as in the covariance matrix) to obtain nonlinear generalizations of classical feature extraction methods.
Material
Hyperspectral image processing (60 hours) - G. Camps-Valls
We introduce the main concepts of hyperspectral image processing. We start by a soft introduction to hyperspectral image processing, the standard processing chain and the current challenges in the field. Then we analyze the current state of the art in several topics: feature extraction, supervised classificaiton, unmixing and abundance estimation and retrieval of biophysical parameters. All the methods and techniques studied are reviewed both theoretically and through MATLAB exercises.
Material
Machine learning and signal processing for remote sensing data analysis (IGARSS'14 tutorial) - G. Camps-Valls and D. Tuia
In this tutorial, we will present the remote sensing image processing chain, and take the attendants on a tour of different strategies for feature extraction, classification, unmixing, retrieval, and pattern analysis for data understanding. On the one hand, we will present powerful methodologies for remote sensing data classification: extracting knowledge from data, including interactive approaches via active learning, classifiers that encode prior knowledge and invariances, semisupervised learning that exploit the information of unlabeled data, and domain adaptation to compensate for shifts in the ever-changing data distributions. On the other hand, we will pay attention to recent advances in bio-geophysical parameter estimation that incorporate heteroscedasticity, online adaptation, and problem understanding. From there we will take a leap towards the more challenging step of understanding the geoscience problems from data by reviewing the latest advances in (directed) graphical models, structure learning and empirical causal inference. Beyond theory, we will also present results of recent studies illustrating all the covered issues. Finally, we will provide code to the attendees to try the different methodologies and provide a solid ground for their future experimentations.
Material
The GLaSS training material builds on the global lakes use cases - Ana B. Ruescas & GLaSS team
 The GLaSS training material builds on the global lakes use cases of GLaSS. It allows students and professionals in fields as aquatic ecology, environmental technology, remote sensing and GIS to learn about the possibilities of optical remote sensing of water quality, by using the Sentinel-2 and Sentinel-3 satellites and Landsat 8.
Material

 A short introducttion to Google Earth Engine.
Material

34
Interesting information / Regression and system identification
« on: October 11, 2019, 04:34:11 PM »
1. simpleR v2.1: simple Regression toolbox

The simple Regression toolbox, simpleR, contains a set of functions in Matlab to illustrate the capabilities of several statistical regression algorithms. simpleR contains simple educational code for linear regression (LR), decision trees (TREE), neural networks (NN), support vector regression (SVR), kernel ridge regression (KRR), aka Least Squares SVM, Gaussian Process Regression (GPR), and Variational Heteroscedastic Gaussian Process Regression (VHGPR). We also include a dataset of collected spectra and associated chlorophyll content to illustrate the training/testing procedures. This is just a demo providing a default initialization. Training is not at all optimized. Other initializations, optimization techniques, and training strategies may be of course better suited to achieve improved results in this or other problems. We just did it in the standard way for illustration and educational purposes, as well as to disseminate these models.

Last version of the toolbox is in GitHub: https://github.com/IPL-UV/simpleR2. Fair Kernel Learning3. MSVR: Multioutput Support Vector Regression

Standard SVR formulation only considers the single-output problem. In the case of several output variables, other methods (neural networks, kernel ridge regression) must be deployed, but the good properties of SVR are lost: hinge-loss function and sparsity. The proposed model M-SVR extends the single-output SVR by taking into account the nonlinear relations between features but also among the output variables, which are typically inter-dependent.

References
Multioutput support vector regression for remote sensing biophysical parameter estimation Tuia, D. and Verrelst, J. and Alonso, L. and Perez-Cruz, F. and Camps-Valls, G. IEEE Geoscience and Remote Sensing Letters 8 (4):804-808, 2011

4. Epsilon-Huber Support Vector Regression5. SS-SVR: Semi-supervised Support Vector Regression6. ARTMO: Automated Radiative Transfer Models Operator

The in-house developed Automated Radiative Transfer Models Operator (ARTMO) Graphic User Interface (GUI) is a software package that provides essential tools for running and inverting a suite of plant RTMs, both at the leaf and at the canopy level. ARTMO facilitates consistent and intuitive user interaction, thereby streamlining model setup, running, storing and spectra output plotting for any kind of optical sensor operating in the visible, near-infrared and shortwave infrared range (400-2500 nm). the ARTMO package includes physical, statistical and hybrid inversion and model emulation. Some modules are pure machine learning techniques for regression, active learning, dimensionality reduction and feature ranking!

References
Toward a semiautomatic machine learning retrieval of biophysical parameters Caicedo, J.P.R. and Verrelst, J. and Munoz-Mari, J. and Moreno, J. and Camps-Valls, G. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 7 (4):1249-1259, 2014

7. KARMA: Kernel AutoRegressive Moving Average with the Support Vector Machine8. ARX-RVM: Autorregressive eXogenous Relevance Vector Machine
9. KSNR: Kernel Signal to Noise Ratio

The kernel signal to noise ratio (KSNR) considers a least squares regression model that maximizes the signal variance while minimizes the estimated noise variance in a reproducing kernel Hilbert space (RKHS). The KSNR can be used in any kernel method to deal with correlated (possibly non-Gaussian) noise. KSNR yields more fitted solutions and extracts more noise-free features when confronted with standard approaches.

References
Learning with the kernel signal to noise ratio Gomez-Chova, L. and Camps-Valls, G. IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2012

Link: https://isp.uv.es/soft_regression.html

35
Interesting information / Remote sensing applications
« on: October 11, 2019, 04:31:30 PM »
1. Randomized Kernels for Large Scale Earth Observation Applications

Kernel methods constitute a family of powerful machine learning algorithms, which have found wide use in remote sensing and geosciences. However, kernel methods are still not widely adopted because of the high computational cost when dealing with large scale problems, such as the inversion of radiative transfer models or the classification of high spatial-spectral-temporal resolution data. This paper introduces an efficient kernel method for fast statistical retrieval of atmospheric and biophysical parameters and image classification problems real applications. We rely on a recently presented approximation to shift-invariant kernels using projections on random Fourier features. The method allows to approximate a kernel matrix with a set of projections on random bases sampled from the Fourier domain. The method is simple, computationally very efficient in both memory and processing costs, and easily parallelizable.

2. MERIS/AATSR Synergy Cloud Screening

A module for the BEAM plaftorm that provides cloud screening within the MERIS/AATSR Synergy Toolbox. The MERIS/AATSR Synergy Toolbox provides processing schemes for improved cloud screening, global aerosol retrieval and land atmospheric correction using the combined multi-spectral and multi-angle information from geo-located and geo-registered MERIS and AATSR measurements.

3. SIMFEAT: A simple MATLAB(tm) toolbox of linear and kernel feature extraction

Toolbox of linear and kernel feature extraction: (1) Linear methods: PCA, MNF, CCA, PLS, OPLS, and (2) Kernel feature extractors: KPCA, KMNF, KCCA, KPLS, KOPLS and KECA.

References
Kernel multivariate analysis framework for supervised subspace learning: A tutorial on linear and kernel multivariate methods Arenas-Garcia, J. and Petersen, K.B. and Camps-Valls, G. and Hansen, L.K. IEEE Signal Processing Magazine 30 (4):16-29, 2013

4. simpleR v2.1: simple Regression toolbox5. simpleUnmix: simple Unmixing and Abundance estimation toolbox6. simpleClass: Simple Classification Toolbox

A set of train-test simple educational functions for data classification: LDA, QDA, Mahalanobis-distance classifier, decision trees, random forests, SVM, Boosting, Bagging, Gaussian process classifiers, etc.

7. ALTB: Active Learning MATLAB(tm) Toolbox8. ARTMO: Automated Radiative Transfer Models Operator

The in-house developed Automated Radiative Transfer Models Operator (ARTMO) Graphic User Interface (GUI) is a software package that provides essential tools for running and inverting a suite of plant RTMs, both at the leaf and at the canopy level. ARTMO facilitates consistent and intuitive user interaction, thereby streamlining model setup, running, storing and spectra output plotting for any kind of optical sensor operating in the visible, near-infrared and shortwave infrared range (400-2500 nm). the ARTMO package includes physical, statistical and hybrid inversion and model emulation. Some modules are pure machine learning techniques for regression, active learning, dimensionality reduction and feature ranking!

References
Toward a semiautomatic machine learning retrieval of biophysical parameters Caicedo, J.P.R. and Verrelst, J. and Munoz-Mari, J. and Moreno, J. and Camps-Valls, G. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 7 (4):1249-1259, 2014

Link: https://isp.uv.es/soft_rs.html

36
Interesting information / Classification, change and anomaly detection
« on: October 11, 2019, 04:28:50 PM »

1. HyperLabelMe: A web platform for benchmarking remote-sensing image classifiers2. ALTB: Active Learning MATLAB(tm) Toolbox3. EC-ACD: Elliptically Contoured Anomaly Change Detection

A simple Toolbox for Anomaly Change Detection (ACD) with Gaussianity assumptions and Elliptically Contoured (EC) distributions.

References
A family of kernel anomaly change detectors Longbotham, N. and Camps-Valls, G. IEEE Whispers, 2015
Robustness analysis of ellipicatlly contoured multi- and hyperspectral change detection algorithms M. A. Belenguer, Longbotham, N. and Camps-Valls, G. Submitted, 2016

4. simpleClass: Simple Classification Toolbox

A set of train-test simple educational functions for data classification: LDA, QDA, Mahalanobis-distance classifier, decision trees, random forests, SVM, Boosting, Bagging, Gaussian process classifiers, etc.

Last version of the toolbox is in GitHub: https://github.com/IPL-UV/simpleClass.

5. Graph kernels for spatio-spectral classification

A graph kernel for spatio-spectral remote sensing image classification with support vector machines (SVM). The method considers higher order relations in the neighborhood (beyond pairwise spatial relations) to iteratively compute a kernel matrix for SVM learning. The proposed kernel is easy to compute and constitutes a powerful alternative to existing approaches.

References
Spatio-spectral remote sensing image classification with graph kernels Camps-Valls, G. and Shervashidze, N. and Borgwardt, K.M. IEEE Geoscience and Remote Sensing Letters 7 (4):741-745, 2010

6. Semi-supervised Graph-based Classification7. BagSVM: Bag Support Vector Machine8. Our modified libSVM

Precomputed kernels, e-Huber cost function, accuracy assessment, and other useful things for SVM methods.

9. Large Margin Filtering SVM10. UKC: Unsupervised kernel change detection

This code+data snippet implements the automatic change detection algorithm presented in the SPIE 2010 and IGARSS 2011 papers. It consists in three steps: initialization (histogram-based), automatic parameter tuning and change map generation, with classical kmeans, gaussian kernel kmeans and by clustering the difference image in the feature spaces. Loops allows to perform different experiments and the user can choose different parameters (number of experiments, kernel function and parameters to search, number of pseudo training samples,...). Moreover, one can easily adapt the code for comparisons to other algorithms and images, as well as adapting the code for personal developments.
References
Unsupervised change detection by kernel clustering Volpi, M. and Tuia, D. and Camps-Valls, G. and Kanevski, M. Proceedings of SPIE - The International Society for Optical Engineering 7830 2010
Unsupervised change detection in the feature space using kernels Volpi, M. and Tuia, D. and Camps-Valls, G. and Kanevski, M. International Geoscience and Remote Sensing Symposium (IGARSS) 2011

Link: https://isp.uv.es/soft_classification.html

37
The following modules are available under this category:

1. DRR: Dimensionality Reduction via Regression

Dimensionality Reduction via Regression (DRR) is a manifold learning technique intended to remove the residual statistical dependence between PCA components due to the curvature in the dataset. DRR is based on the prediction of PCA coefficients from neighbor coefficients using multivariate regression (hence generalizing PPA). DRR is a computationallly convenient step forward to SPCA in the family of manifold learning techniques generalizing PCA through the use of curves instead of stright lines (that includes NL-PCA, PPA, and SPCA).

References
Dimensionality reduction via regression in hyperspectral imagery Laparra, V. and Malo, J. and Camps-Valls, G. IEEE Journal on Selected Topics in Signal Processing 9 (6):1026-1036, 2015

2. PPA: Principal Polynomial Analysis3. SIMFEAT: A simple MATLAB(tm) toolbox of linear and kernel feature extraction

Toolbox of linear and kernel feature extraction: (1) Linear methods: PCA, MNF, CCA, PLS, OPLS, and (2) Kernel feature extractors: KPCA, KMNF, KCCA, KPLS, KOPLS and KECA.

Last version of the toolbox is in GitHub: https://github.com/IPL-UV/simFeat.

References
Kernel multivariate analysis framework for supervised subspace learning: A tutorial on linear and kernel multivariate methods Arenas-Garcia, J. and Petersen, K.B. and Camps-Valls, G. and Hansen, L.K. IEEE Signal Processing Magazine 30 (4):16-29, 2013

4. RBIG: Rotation-Based Iterative Gaussianization5. SPCA: Sequential Principal Curves Analysis6. SSKPLS: Semisupervised Kernel Partial Least Squares7. HOCCA: Higher Order Canonical Correlation Analysis8. SSMA: SemiSupervised Manifold Alignment

The SSMA Toolbox is a Matlab Toolbox for the semisupervised manifold alignment of generic data without the need of having corresponding pairs, just a set of (few) labeled samples in each domain.

References
Semisupervised manifold alignment of multimodal remote sensing images Tuia, D. and Volpi, M. and Trolliet, M. and Camps-Valls, G. IEEE Transactions on Geoscience and Remote Sensing 52 (12):7708-7720, 2014

9. KEMA: Kernel Manifold Alignment

Kernelization of SSMA, which allows for much better semantic alignments of multisource data.

References
Kernel Manifold Alignment for Domain Adaptation Tuia, D. and G. Camps-Valls PLoS ONE, 2016

10. OKECA: Optimized Kernel Entropy Component Analysis11. KSNR: Kernel Signal to Noise Ratio

The kernel signal to noise ratio (KSNR) considers a feature extraction model that maximizes the signal variance while minimizes the estimated noise variance in a reproducing kernel Hilbert space (RKHS). The KSNR can be used for dimensionality reduction as an excellent alternative to kPCA when dealing with correlated (possibly non-Gaussian) noise. KSNR yields more fitted solutions and extracts more noise-free features when confronted with standard approaches.

References
Learning with the kernel signal to noise ratio Gomez-Chova, L. and Camps-Valls, G. IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2012

12. EPLS: Unsupervised sparse convolutional neural networks for feature extraction

EPLS stands for Enhancing Population and Lifetime Sparsity, a very good alternative to achieve sparse representations when training machines such as convolutional neural nets. EPLS is a meta-parameter free, off-the-shelf, simple and fast unsupervised feature learning algorithm, which exploits a new way of optimizing for sparsity. The algorithm sets an output target with one "hot code" while ensuring a strong form of lifetime sparsity to avoid dead outputs and optimizes for that specific target to learn the dictionary bases.

References
Unrolling loopy top-down semantic feedback in convolutional deep networks Carlo Gatta, Adriana Romero, Joost van de Weijer. Deep-vision workshop CVPR, 2014.
Unsupervised Deep Feature Extraction Of Hyperspectral Images Adriana Romero, Carlo Gatta, Gustavo Camps-Valls IEEE Workshop on Hyperspectral Image and Signal Processing, Whispers, 2014
Unsupervised Deep Feature Extraction for Remote Sensing Image Classification Romero, A. and Gatta, C. and Camps-Valls, G. Geoscience and Remote Sensing, IEEE Transactions on 2015
Shared feature representations of LiDAR and optical images: Trading sparsity for semantic discrimination Campos-Taberner, M. and Romero, A. and Gatta, C. and Camps-Valls, G. Geoscience and Remote Sensing Symposium (IGARSS), 2015 IEEE International, 2015

Link: https://isp.uv.es/soft_feature.html

38
Hydrological sciences / RSPARROW: A Water Quality Model using R
« on: October 10, 2019, 10:27:08 PM »
RSPARROW is a system of R scripts and functions for executing and evaluating SPARROW models that generates graphical, map, and tabular output. Users operate the system within RStudio from a single control script that accesses the supporting input files and functions. Only minimal knowledge of R is required to use the system.
SPARROW (SPAtially Referenced Regressions on Watershed attributes) is a spatially explicit, hybrid (statistical and mechanistic) water-quality model developed by the USGS. The model has been used to quantify the sources and transport of contaminants in watersheds of widely varying sizes, from catchment to continental scales. SPARROW includes three major process components that explain spatial variability in stream water quality:  (1) contaminant source generation, (2) land-to-water delivery, and (3) stream and reservoir transport and decay. The non-linear and mechanistic structure of the model includes mass balance constraints and non-conservative transport components. This includes factors that control the attenuation and delivery of contaminants to streams via surficial and subsurface pathways and the removal of contaminants in streams and reservoirs, according to first-order decay kinetics. SPARROW is structured as a network of one-dimensional stream segments and their contributing drainage areas.
The RSPARROW documentation describes the steps necessary to estimate the static version of the model. The static model provides reach-level predictions (and uncertainties) of the long-term mean annual water-quality loads, yields, and flow-weighted concentrations. The predictions also include the shares of the load attributable to individual upstream sources and predictions of the mass quantities of the total load and individual sources that are delivered to downstream water bodies.


Link: https://code.usgs.gov/water/stats/rsparrow

39
Land-cover (LC) is defined as the observed physical cover on the Earth surface. Objects and surfaces are segmented into classes. Their number, type, and definition (nomenclature) vary with the application and the geographical scale. LC description is the core information layer for many interdisciplinary scientific and environmental studies. Accurate and up-to-date maps over large areas are mandatory baselines. A large number of public policies, in particular in the European Union, are driven by such knowledge: climate change mitigation, reduction of risks and threats, global sustainable development.

Remote sensing through automatic Earth-Observation image analysis has been widely recognized as the most feasible approach to derive LC information over large areas. It is now accepted that manual or semi-automatic generation of LC geodatabases through visual inspection of EO images is not a sustainable answer for current needs. Such issue is exacerbated with the increasing demand for semantic and geometric accuracy, and updateness.

https://lslc.sciencesconf.org/

40
AbstractKey points
    Earth and planetary landscapes are created by the erosion and deposition of particulate material; this discipline is called geomorphology.

    Soil, rocks and ice relax over geologic timescales, but may also fluidize under shear or lubrication; thus, glassy dynamics, rigidity transitions and rheology are central concepts.

    Progress in soft-matter physics can be extended to improve the understanding of geophysical flows that shape landscapes.

    Landscapes present a wider range of material heterogeneity, system geometry and excitations than have been examined in physics experiments, presenting new challenges and opportunities.

    Soft-matter physics and geomorphology are long-lost relatives, and we outline promising avenues for reunification and collaboration.

https://www.nature.com/articles/s42254-019-0111-x

41
Groundwater flow drives partitioning

Soil evaporation and plant transpiration together contribute a substantial proportion of terrestrial freshwater fluxes. Land surface models are used to understand the partitioning of these fluxes on a continental scale; however, model outputs are often inconsistent with stable isotope observations. Maxwell and Condon incorporated dynamic groundwater flow into an integrated hydrologic model simulation for the entire United States. The model showed that water table depth and lateral flow strongly affect transpiration partitioning, thus explaining the inconsistencies between observations and models.

Abstract:https://science.sciencemag.org/content/353/6297/377.full?ijkey=oMiU7ksao/GBE&keytype=ref&siteid=sci

42
Interesting information / Let's Talk About Water Film Festival
« on: October 06, 2019, 09:59:35 PM »
?
Whether you call yourself a hydrologist, climatologist, ecologist, geographer, social scientist, artist, or knowledge keeper, you can submit a film about water, nature, and humans. The film may link to more in-depth information in any medium (e.g., a web page, blog, podcast). The target audience is the general public, their leaders, and people who are engaged in water stewardship and management. Your film should connect to people, inspiring them to take action or helping them understand the world more deeply.

https://www.waterfilmprize.com/

43
Interesting information / Numba: A high performance Python compiler
« on: October 06, 2019, 09:31:33 PM »
Numba translates Python functions to optimized machine code at runtime using the industry-standard LLVM compiler library. Numba-compiled numerical algorithms in Python can approach the speeds of C or FORTRAN.
You don't need to replace the Python interpreter, run a separate compilation step, or even have a C/C++ compiler installed. Just apply one of the Numba decorators to your Python function, and Numba does the rest.

http://numba.pydata.org/

45
Interesting information / xESMF: Universal Regridder for Geospatial Data
« on: October 05, 2019, 12:17:53 PM »
xESMF is a Python package for regridding. It is
 
  • Powerful: It uses ESMF/ESMPy as backend and can regrid between general curvilinear grids with all ESMF regridding algorithms, such as bilinear, conservative and nearest neighbour.
  • Easy-to-use: It abstracts away ESMF's complicated infrastructure and provides a simple, high-level API, compatible with xarray as well as basic numpy arrays.
  • Fast: It is faster than ESMPy's original Fortran regridding engine in serial case, and also supports dask for out-of-core, parallel computation.
Please see online documentation, or play with example notebooks on Binder.

Pages: 1 2 [3] 4 5 ... 29