Spectral Unmixing

This study presents a benchmark for the detection of bare soil from EnMAP hyperspectral imagery, addressing the challenge of mixed pixels, a known limitation in satellitebased soil property estimation. A reference bare soil mask was constructed using a histogram separation method applied to spectral indices of a multitemporal Sentinel-2 composite. Using this higher resolution reference, we evaluate two classification approaches on EnMAP data: spectral index thresholding and linear spectral unmixing. The latter achieves higher classification performance, indicating improved reliability under heterogeneous surface conditions. The benchmark dataset created for this study is publicly available 1 .

100 quadrillion operations per second. $76,000 prototype. 530 watts. The PGP-X is a free-space optical accelerator that performs matrix multiplication at the speed of light—faster than any electronic processor ever built. Using a DMD, diffractive optical element, and high-speed camera, it achieves sub-millisecond latency on dense linear algebra. This document provides complete build specifications: optical layout, component selection, assembly procedures, calibration methods, and software API. All parts commercially available. Based on 6DUFT. Open hardware, patent-pending.

Mitigating the impact of mining activity is a significant challenge for environmental management. New approaches are required based on environmental information from in-situ, airborne and satellite observations. Airborne hyperspectral remote sensing and geophysical techniques are two such approaches being pursued at the BGS to characterise mine waste and its host rocks and then to monitor its dispersal in the environment. This paper will report on two projects that have followed on from the successful completion of the MINEO project. MINEO applied hyperspectral remote sensing to a range of active and abandoned mining environments across Europe. Parys Mountain, on Anglesey, was the first site under investigation using HyMap data following MINEO. Here, the main task has been mapping the different tailings and characterising them mineralogically. Secondary investigations have involved the study of vegetation stress around the mine. Low-level airborne geophysical data already exist for the Bolsover area from the HiRES-1 survey in 1998. HyMap data acquired during the NERC SAR and Hyperspectral Campaign (SHAC) over the Shirebrook mine within the HiRES-1 area have been used to evaluate the use of airborne geophysical and hyperspectral data together to map open cast mines in the region. Mine site classifications were developed based on both field and image spectral interpretations. In all these projects, various techniques have been applied in order to classify mine sites in terms of mineralogy and contamination. The approach adopted is to use MNF and PPI approaches to extract end-member spectra and then classify the data using the SAM algorithm.

High-speed 3D optical microscopy is an indispensable requirement in studying rapid processes like signaling in neuronal networks, flagellar motion, or complex motion of living and highly dynamic subcellular components. However, the 3D imaging capability often compromises either the acquisition speed or the complexity of the imaging system. Multi-plane imaging offers a parallelized acquisition of several focal planes at the same time, which tremendously enhances the temporal resolution. Nonetheless, most of the current multi-plane approaches are complex and require several optical elements to create multiple beams with different optical path lengths, limiting their broad application. To address these problems, in this thesis, employing a multi-plane prism, we introduce three novel techniques that each improve the concept of rapid 3D optical microscopy in a different direction: First, we accomplish a multi-plane detection for a phase-contrast microscope resulting in a 3D, label-free and non-complex imaging system with a ∼ 4ms of temporal resolution. We applied this system in real-time studying isolated bare axonemes of Chlamydomonas beating in the vicinity of a surface. This enabled us to observe their non-zero torsional motion where the torsion sign slowly changes from negative at the basal end toward positive at the distal end of flagella. Second, straightforwardly, we combine an experimental spectral unmixing setup with the multi-plane detection and obtain an instant 3color 3D microscope with a minimal axial color aberration of 140 nm in a ∼ 2.5 µm axial range. The performance of such a microscope is verified in volumetric imaging of three different subcellular components in fixed COS-7 cells. Third, a single-color multi-plane fluorescence microscope has been employed as a rapid 3D particle tracking velocimetry method in microscale to study single particles tracing the so-called Marangoni flow. Using that, besides the velocity fields of single particles, we achieve a localization precision of ∼ 60 nm in the lateral and ∼ 300 nm in the axial direction over a 120 × 120 × 13 µm 3 field of view. Altogether, using these three techniques, we extend the temporal resolution, throughput, and applicability of 3D optical microscopy. In another project, we implemented image scanning microscopy simultaneously in two colors on a conventional wide-field microscope by applying an LED light source and a digital micro-mirror device. This permits us to realize an inexpensive, speckle noise-free, and easy-to-implement super-resolution add-on for a conventional epifluorescence microscope.

Proximal soil sensing (PSS) is a promising approach when it comes to detailed characterization of spatial soil heterogeneity. Since none of existing PSS systems can measure all soil information needed for implementation precision agriculture, sensor data fusion can provide a reasonable al- ternative to characterize the complexity of soils. In this study, we fused the data measured using a gamma-ray sensor, an apparent electrical conductivity (ECa) sensor, and a commercial Veris MSP3 platform including a optical sensor measuring soil reflectance at 660 nm and 940 nm, a soil ECa sensor and a pH sensor, with the addition of topography for the prediction of several soil properties, i.e. soil organic matter, pH, buffer pH, phosphorus, potassium, calcium, magnesium, aluminum. A total of 65 sampling locations were selected from a 38.5 ha field in Ontario, Canada. Among them, 35 locations were selected by a random stratified sampling strategy. The stratification grid was 1 ha. Sampling was ...

Hyperspectral unmixing (HU) is crucial for extracting material information from hyperspectral images (HSI) obtained through remote sensing. Although linear unmixing methods are widely used due to their simplicity, they only address linear mixing effects. Nonlinear mixing models, while more complex, often focus solely on the nonlinear aspects affecting individual pixels. However, in practice, light reflected from materials within a pixel experiences linear and nonlinear interactions, necessitating a hybrid mixing model (HMM) that leverages spatial and spectral information. This work proposes a novel deep learning-based autoencoder (AE) with dual-stream decoders to enhance spectral unmixing. Our approach employs multitask learning (MTL) to process spatial and spectral information concurrently. Specifically, one decoder stream performs linear unmixing from HSI patches, while the other stream utilizes fully connected layers to capture and model the nonlinear interactions within the data. By integrating linear and nonlinear information, our method improves the accuracy of unmixing the mixed spectrum. We validate the effectiveness of our architecture on three real-world HSI datasets and compare its performance against various baseline methods. Experimental results consistently demonstrate that our approach outperforms existing methods, as evidenced by superior spectral angle distance (SAD) and mean squared error (MSE) metrics.

In recent years, nonlinear unmixing of hyperspectral data has become an attractive topic in hyperspectral image analysis, because nonlinear models appear as more appropriate to represent photon interactions in real scenes. For this challenging problem, nonlinear methods operating in reproducing kernel Hilbert spaces have shown particular advantages. In this paper, we derive an efficient nonlinear unmixing algorithm based on a recently proposed linear mixture/nonlinear fluctuation model. A multi-kernel learning support vector regressor is established to determine material abundances and nonlinear fluctuations. Moreover, a low complexity locally-spatial regularizer is incorporated to enhance the unmixing performance. Experiments with synthetic and real data illustrate the effectiveness of the proposed method.

Pattern recognition has as its objective to classify objects into different categories and classes. It is a fundamental component of artificial intelligence and computer vision. This thesis investigates the application of an efficient optimization method, known as Particle Swarm Optimization (PSO), to the field of pattern recognition and image processing. First a clustering method that is based on PSO is proposed. The application of the proposed clustering algorithm to the problem of unsupervised classification and segmentation of ...

We address the problem of complexity reduction in hyperspectral image unmixing. When the hyperspectral images are highly resoluted, we propose to select a limited number of pixels, therefore reducing dramatically the size of the data. Then, the related mixtures are used as inputs to a positive source separation algorithm. Our pixel selection procedure is based on a convex cone analysis of the data mixtures; indeed, positive mixtures of sources are embedded in a convex cone whose boundary contains complete available information regarding the sources. We search for the least number of mixtures embedding the convex cone and then store the corresponding pixel indices as the selected pixels. We apply this method to the analysis of hyperspectral images of bacterial cells obtained on a confocal microscope. The bacterial cells, acting as whole-cell biosensors, display great potential as living transducers in sensing applications.

Geometrical registration of two images is nowadays a current and important step in remote sensing in view of further processing and interpretation of the data. Therefore, geometrical registration of images with different ground resolutions is useful for a better comprehension of dynamic processes [i.e., deforestation, desertijlcation), as well as for extrapolating models of interpretation from a small region to larger areas. We present a new method of automatic registration, based on a multiresolution decomposition of the images using the wavelet transform. The main properties of this transform, as well as how the use of the wavelet model leads to an automated geometric matching between the different images, are described. Our technique has been applied to diflerent sets of data acquired with the same sensors [SPOT HRV; Landsat MSS) as well as from different sensors with different resolutions (SPOT HRV and Landsat MSS; Landsat TM and SPOT H R ~.

The spaceborne ESA-mission CHRIS-PROBA (Compact High Resolution Imaging Spectrometer-Project for On-board Autonomy) provides hyperspectral and multi-angular data of selected terrestrial targets ). For vegetated surfaces, the spectral information content of CHRIS data may yield the biochemical and biophysical properties of a vegetation canopy, while the directional component may deliver additional information on its canopy structure. Here, the multi-angular CHRIS observations are investigated and related to surface structure using the parametric Rahman-Pinty-Verstraete (RPV) model. RPV parameters, in particular the Minnaert function parameter k, are able to quantify the anisotropy of surface reflectance, which in turn is sensitive to the structure and heterogeneity of the observed surface. Thus to investigate the potential of multi-angular high spatial resolution data for delivering quantitative surface structure information the retrieved values of the k parameter are compared with proxies derived from detailed LIDAR observations describing the 3-D canopy structure of the target site. Based on the achieved results important factors affecting the anisotropy interpretation are identified and discussed.

In this letter, we discuss the use of multicore processors in the acceleration of endmember extraction algorithms for hyperspectral image unmixing. Specifically, we develop computationally efficient versions of two popular fully automatic endmember extraction algorithms: orthogonal subspace projection and N-FINDR. Our experimental results, based on the analysis of hyperspectral data collected by the National Aeronautics and Space Administration Jet Propulsion Laboratory's Airborne Visible InfraRed Imaging Spectrometer, indicate that endmember extraction algorithms can significantly benefit from these inexpensive high-performance computing platforms, which can offer real-time response with some programming effort.

Pesticide residue analysis of agricultural produce is vital because of associated health concerns, highlighting the need for effective non-destructive techniques. This study introduces a method that combines short-wavelength infrared hyperspectral imaging with spectral unmixing to detect chlorfenapyr and azoxystrobin residues on perilla leaves. Sixty-six leaves were treated with pesticides at concentrations between 0 and 0.06%. The study utilized multicurve resolution-alternating least squares (MCR-ALS), a spectral unmixing method, to identify and visualize the distribution of pesticide residues. This technique achieved lack-of-fit values of 1.03% and 1.78%, with an explained variance of 99% for both pesticides. Furthermore, a quantitative model was developed that integrates insights from MCR-ALS with Gaussian process regression to estimate chlorfenapyr residue concentrations, resulting in a root mean square error of double cross-validation (RMSEV) of 0.0012% and a double cross-validation coefficient of determination (R2v) of 0.99. Compared to other chemometric approaches, such as partial least squares regression and support vector regression, the proposed integrated method decreased RMSEV by 67.57% and improved R2v by 2.06%. The combination of hyperspectral imaging with spectral unmixing offers advancements in the real-time monitoring of agricultural product safety, supporting the delivery of high-quality fresh vegetables to consumers.

In this letter, we discuss the use of multicore processors in the acceleration of endmember extraction algorithms for hyperspectral image unmixing. Specifically, we develop computationally efficient versions of two popular fully automatic endmember extraction algorithms: orthogonal subspace projection and N-FINDR. Our experimental results, based on the analysis of hyperspectral data collected by the National Aeronautics and Space Administration Jet Propulsion Laboratory's Airborne Visible InfraRed Imaging Spectrometer, indicate that endmember extraction algorithms can significantly benefit from these inexpensive high-performance computing platforms, which can offer real-time response with some programming effort.

It is now possible to collect hyperspectral video sequences (HVS) at a near real-time frame rate. The wealth of spectral, spatial and temporal information of those sequences is particularly appealing for chemical gas plume tracking. Existing state-of-the-art methods for such applications however produce only a binary information regarding the position and shape of the gas plume in the HVS. Here, we introduce a novel method relying on spectral unmixing considerations to perform chemical gas plume tracking, which provides information related to the gas plume concentration in addition to its spatial localization. The proposed approach is validated and compared with three state-of-the-art methods on a real HVS.

Multiple subpixel shifted images (MSIs) from the same area can be incorporated to improve the accuracy of softthen-hard subpixel mapping (STHSPM). In this paper, a novel method that derives higher resolution MSIs with more spatialspectral information (MSI-SS) is proposed. First, coarse MSIs produce two high-resolution MSIs for each class respectively by two paths at the same time. The spatial path produces the highresolution MSIs by soft classification followed by interpolation. And, the other high-resolution MSIs are derived from the spectral path by interpolation followed by soft classification. Then the higher resolution MSIs with more spatial-spectral information for each class are derived by integrating the aforementioned two kinds of high-resolution MSIs by the appropriate weight. Finally, the integrated higher resolution MSIs for each class are used to allocate hard class labels to subpixels. The proposed method is fast and takes more spatial-spectral information of the original MSIs into account. Experiments on three real hyperspectral remote sensing images show that the proposed method produce higher SPM accuracy result.

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

The characterization of snow extent is critical for a wide range of applications. Since 1966, snow maps at different spatial resolutions have been produced using various satellite sensor images. Nowadays, the most widely used products are likely those derived from Moderate-Resolution Imaging Spectroradiometer (MODIS) data, which cover the whole Earth at a near-daily frequency. There are a variety of snow mapping methods for MODIS data, based on different methodologies and applied at different spatial resolutions. Up to now, all these products have been tested and evaluated separately. This study aims to compare the methods currently available for retrieving snow from MODIS data. The focus is on fractional snow cover, which represents the snow cover area at the subpixel level. We examine the two main approaches available for generating such products from MODIS data; namely, linear regression of the Normalized Difference Snow Index (NDSI) and spectral unmixing (SU). These two approaches have resulted in several methods, such as MOD10A1 (the NSIDC MODIS snow product) for NDSI regression, and MODImLAB for SU. The assessment of these approaches was carried out using higher resolution binary snow maps (i.e., showing the presence or absence of snow) at spatial resolutions of 10, 20, and 30 m, produced by SPOT 4, SPOT 5, and LANDSAT-8, respectively. Three areas were selected in order to provide landscape diversity: the French Alps (117 dates), the Pyrenees (30 dates), and the Moroccan Atlas (24 dates). This study investigates the impact of reference maps on accuracy assessments, and it is suggested that NDSI-based high spatial resolution reference maps advantage NDSI medium-resolution snow maps. For MODIS snow maps, the results show that applying an NDSI approach to accurate surface reflectance corrected for topographic and atmospheric effects generally outperforms other methods for the global retrieval of snow cover area. The improvements to the newer version of MOD10A1 (Collection 6) compared to the older version (Collection 5) are significant. Products based on SU provide a good alternative and more accurate retrieval of the snow fraction where wider ranges of land covers are concerned. The fusion process and its resulting 250 m spatial resolution product improve snow line retrieval. False detection in mixed pixels, probably due to the spectral variability associated with the various materials in the spectral mixture, has been identified as an area that will require improvement.

Hyperspectral image unmixing is an inverse problem aiming at recovering the spectral signatures of pure materials of interest (called endmembers) and estimating their proportions (called abundances) in every pixel of the image. However, in spite of a tremendous applicative potential and the avent of new satellite sensors with high temporal resolution, multitemporal hyperspectral unmixing is still a relatively underexplored research avenue in the community, compared to standard image unmixing. In this paper, we propose a new framework for multitemporal unmixing and endmember extraction based on a state-space model, and present a proof of concept on simulated data to show how this representation can be used to inform multitemporal unmixing with external prior knowledge, or on the contrary to learn the dynamics of the quantities involved from data using neural network architectures adapted to the identification of dynamical systems.

The tensor-based anomaly detection (AD) model has attracted increasing interest in the hyperspectral image (HSI) community. Since it is powerful in maintaining spatial and spectral structures, an HSI is essentially a third-order tensor. In this article, we propose a novel AD method based on a lowrank background linear mixing model of the scene background. The obtained abundance maps possess more distinctive features than the raw data, which is beneficial for identifying an anomaly from the background. Specifically, the low-rank tensor background is approximated as the mode-3 product of an abundance tensor and endmember matrix. Due to the spatial sparse and smooth natures of abundance maps, the ℓ 1 -norm is introduced to enforce sparseness, and the total variation (TV) regularizer is adopted to encourage spatial smoothness. Moreover, the typical great correlation among abundance vectors implies the low-rank structure of abundance maps. Compared with the rigorous lowrank constraint, a soft low-rank regularization is imposed on the background in order to leverage its spatial homogeneity. Its strictness is controlled by scalar parameters. For the anomaly part, the anomaly spectra account for a small part of the whole scene, and therefore, an ℓ 1,1,2 -norm is defined to characterize its tube-wise sparsity. Notably, Gaussian noise is integrated into the model to suppress confusion with the anomaly. The experimental results on five real datasets demonstrate the outstanding performance of our proposed method. Index Terms-Anomaly detection, low-rank background linear mixing model, mode-3 product, third-order abundance tensor, TV regularizer, ℓ 1,1,2 -norm.

The characterization of snow extent is critical for a wide range of applications. Since 1966, snow maps at different spatial resolutions have been produced using various satellite sensor images. Nowadays, the most widely used products are likely those derived from Moderate-Resolution Imaging Spectroradiometer (MODIS) data, which cover the whole Earth at a near-daily frequency. There are a variety of snow mapping methods for MODIS data, based on different methodologies and applied at different spatial resolutions. Up to now, all these products have been tested and evaluated separately. This study aims to compare the methods currently available for retrieving snow from MODIS data. The focus is on fractional snow cover, which represents the snow cover area at the subpixel level. We examine the two main approaches available for generating such products from MODIS data; namely, linear regression of the Normalized Difference Snow Index (NDSI) and spectral unmixing (SU). These two approaches have resulted in several methods, such as MOD10A1 (the NSIDC MODIS snow product) for NDSI regression, and MODImLAB for SU. The assessment of these approaches was carried out using higher resolution binary snow maps (i.e., showing the presence or absence of snow) at spatial resolutions of 10, 20, and 30 m, produced by SPOT 4, SPOT 5, and LANDSAT-8, respectively. Three areas were selected in order to provide landscape diversity: the French Alps (117 dates), the Pyrenees (30 dates), and the Moroccan Atlas (24 dates). This study investigates the impact of reference maps on accuracy assessments, and it is suggested that NDSI-based high spatial resolution reference maps advantage NDSI medium-resolution snow maps. For MODIS snow maps, the results show that applying an NDSI approach to accurate surface reflectance corrected for topographic and atmospheric effects generally outperforms other methods for the global retrieval of snow cover area. The improvements to the newer version of MOD10A1 (Collection 6) compared to the older version (Collection 5) are significant. Products based on SU provide a good alternative and more accurate retrieval of the snow fraction where wider ranges of land covers are concerned. The fusion process and its resulting 250 m spatial resolution product improve snow line retrieval. False detection in mixed pixels, probably due to the spectral variability associated with the various materials in the spectral mixture, has been identified as an area that will require improvement.

Multiple subpixel shifted images (MSIs) from the same area can be incorporated to improve the accuracy of softthen-hard subpixel mapping (STHSPM). In this paper, a novel method that derives higher resolution MSIs with more spatialspectral information (MSI-SS) is proposed. First, coarse MSIs produce two high-resolution MSIs for each class respectively by two paths at the same time. The spatial path produces the highresolution MSIs by soft classification followed by interpolation. And, the other high-resolution MSIs are derived from the spectral path by interpolation followed by soft classification. Then the higher resolution MSIs with more spatial-spectral information for each class are derived by integrating the aforementioned two kinds of high-resolution MSIs by the appropriate weight. Finally, the integrated higher resolution MSIs for each class are used to allocate hard class labels to subpixels. The proposed method is fast and takes more spatial-spectral information of the original MSIs into account. Experiments on three real hyperspectral remote sensing images show that the proposed method produce higher SPM accuracy result.

The snow coverage area (SCA) is one of the most important parameters for cryospheric studies. The use of remote sensing imagery can complement field measurements by providing means to derive SCA with a high temporal frequency and covering large areas. Images acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) are perhaps the most widely used data to retrieve SCA maps. Some MODIS derived algorithms are available for subpixel SCA estimation, as MODSCAG and MODImLab. Both algorithms make use of spectral unmixing techniques using a fixed set of snow, rocks and other materials spectra (endmembers). We aim to improve the performance of a modified version of MODIm-Lab algorithm by exploring advanced spectral unmixing techniques. Furthermore, we make use of endmember induction algorithms to obtain the endmembers from the data itself instead of using a fixed spectral library. We validate the proposed approach on a case study in the mountainous region of the Alps.