PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303101 (2024) https://doi.org/10.1117/12.3037009
This PDF file contains the front matter associated with SPIE Proceedings Volume 13031, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303102 (2024) https://doi.org/10.1117/12.3013942
This talk describes investigations of shrinkage parameter estimates for covariance matrices used in spectral processing of remote sensing imagery, such as for target, anomaly, or change detection. These estimates are derived in the context of cross-validated fiting of Gaussian likelihood models to the non-target background distribution. Here, the utility of these estimates is evaluated for Gaussian and non-Gaussian distributions. An alternative criterion, based on matched-filter detection of “generic” targets, is derived and compared to the estimated likelihood criterion as a way to choose the shrinkage parameter.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303103 (2024) https://doi.org/10.1117/12.3013476
In this paper, we develop a nested chi-squared likelihood ratio test for selecting among shrinkage-regularized covariance estimators for background modeling in hyperspectral imagery. Critical to many target and anomaly detection algorithms is the modeling and estimation of the underlying background signal present in the data. This is especially important in hyperspectral imagery, wherein the signals of interest often represent only a small fraction of the observed variance, for example when targets of interest are subpixel. This background is often modeled by a local or global multivariate Gaussian distribution, which necessitates estimating a covariance matrix. Maximum likelihood estimation of this matrix often overfits the available data, particularly in high dimensional settings such as hyperspectral imagery, yielding subpar detection results. Instead, shrinkage estimators are often used to regularize the estimate. Shrinkage estimators linearly combine the overfit covariance with an underfit shrinkage target, thereby producing a well-fit estimator. These estimators introduce a shrinkage parameter, which controls the relative weighting between the covariance and shrinkage target. There have been many proposed methods for setting this parameter, but comparing these methods and shrinkage values is often performed with a cross-validation procedure, which can be computationally expensive and highly sample inefficient. Drawing from Bayesian regression methods, we compute the degrees of freedom of a covariance estimate using eigenvalue thresholding and employ a nested chi-squared likelihood ratio test for comparing estimators. This likelihood ratio test requires no cross-validation procedure and enables direct comparison of different shrinkage estimates, which is computationally efficient.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303104 (2024) https://doi.org/10.1117/12.3013806
To explore system sensitivities in hyperspectral subpixel target detection, multivariate methods are applied to output detection metrics generated from a statistical target detection model. The Forecasting and Analysis of Spectroradiometric System Performance (FASSP) statistical model generates probabilities of detection (PD) and false alarm (PFA) using spectral libraries of target and background materials. This allows for the computation of the area under the receiver operating characteristic curve (AUC). To explore sensitivities within elements (e.g. scene, atmosphere, sensor) of the remote sensing system, ensembles of model-based scenarios are generated using combinations of the aerosol visibility, solar angle, and sensor viewing angle. Output detection metrics (PD, AUC) from these scenarios were cached into a high-dimensional tensor, before utilizing multivariate methods (e.g. interpolation and regression) to explore sensitivities and correlations between system variables and detection. Inferences on limitations of detection within the system are drawn from multivariate contour regions which characterize joint parametric parameters required to exceed a desired threshold of detection. The outlined methods aim to provide an initial framework to investigate both specific and generalizable limitations of detection across various scenes (e.g. rural, urban, maritime, and desert), environmental conditions (e.g. solar angle, haze, clouds), sensor characteristics (e.g. noise, viewing angle) and processing configurations (e.g. feature selection, detector algorithm).
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303105 https://doi.org/10.1117/12.3032157
Multispectral and hyperspectral imagery have been used in the remote sensing community for over 30 years, with advances in both instrumentation and analysis algorithms producing significant improvements in capabilities and the breadth of applications. These systems have largely been deployed on manned aircraft and satellites, and now are commonly operated on small Unmanned Aerial Vehicles (UAVs). Recently, with the miniaturization of these systems (and corresponding decrease in cost), they are being deployed in laboratory settings for a wide range of sensing problems. Here, we use both multispectral (MSI) and hyperspectral imaging (HSI) to shed light on cultural heritage artifacts - manuscripts, artwork, maps, and other objects of historical and cultural significance. Multispectral imaging systems are generally used for more qualitative work - particularly enhancing features on artifacts, such as faded or erased text, that are not visible to the naked eye. On the other hand, HSI systems produce data that can be more quantitatively exploited, extracting information about the chemistry of pigments, inks, substrates, and other materials used in the creation of the objects. This information helps historians to understand the object’s codicology, knowledge of how the object was made and potentially modified. Here, I will present a brief overview of the systems being used, as well as show some results for both MSI and HSI systems on medieval maps and manuscripts imaged at Durham University, the Bodleian Library (Univ. of Oxford), as well as other institutions around the world.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303106 (2024) https://doi.org/10.1117/12.3013687
Both MWIR and LWIR wavelength ranges have for long time mostly used for research activities. During recent years industrial use of these wavelength band ranges has also evolved and the same trend is expected to continue in future. Important application areas relate to mineral mapping and recognition, metal industry and black plastics sorting for instance. Key parameters in industrial use, in addition to good performance characteristics and data quality, are the cost and usability of the camera. The Specim hyperspectral MWIR and LWIR cameras employ push-broom imaging spectrograph, with transmission grating and on-axis optics. The cameras have thermally stabilized optics and cryogenically cooled MCT detector with Stirling cooler. The performance of the newly developed MWIR camera is targeted on reflectance measurements with illumination. The performance of hyperspectral camera enables reliable measurement of low reflectance level targets illuminated with moderate temperature heat source, and with less than 10% reflectance with 650°C illumination and frame rates of 380 frames /sec with 154 bands and 640 spatial samples. The SNR of 500 of the new LWIR hyperspectral camera suits emission measurements of normal room temperature targets but is applicable to reflectance measurement with illumination as well. High performance emission measurement with about 150 bands, 640 spatial samples and more than 300 frames per second can be achieved. The performance is verified with testing of several camera units and supported with simulation results. The performance characteristics of NESR and expected SNR with actual measurement parameters are presented.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303107 (2024) https://doi.org/10.1117/12.3014012
This paper presents a machine-learning-informed optimization approach for designing the most cost-effective multispectral system capable of detecting any arbitrarily selected set of materials. The approach presented accepts from the user a list of entities that need to be detected; it then outputs (a) a short list of band centers and bandwidths required for detecting the entities of interest as well as (b) a collection of trained machine-learning models capable of performing those detections with high accuracy. This approach has the potential to help identify cost savings during the design process by allowing proposed hyperspectral systems to be replaced by bespoke multispectral ones – thereby reducing overall mission costs without sacrificing mission performance. A hypothetical design study demonstrates how the proposed approach can automatically design a six-band multispectral system whose detection capabilities are nearly indistinguishable from those of an 80-band hyperspectral system. More precisely, the design procedure was able to reduce the number of required bands by over 90% while only seeing a 0.5% decrease in the average F1 score of a set of machine-learning models trained to identify 26 polymeric materials of interest.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303108 (2024) https://doi.org/10.1117/12.3013143
The Air Force’s Rapid Airfield Damage Assessment (RADA) process was conceived as a means of evaluating airfield pavement assets after attacks to inform subsequent threat mitigation and repair efforts. The classification and geolocation of small objects of interest (< 7.5cm), like unexploded ordnance, is a critical component of this assessment process. In its original form, RADA was conducted manually, exposing teams of service members to dangerous and unknown conditions for hours at a time. In an effort to both expedite and remotely automate this critical task, researchers are developing small Uncrewed Aerial Systems (sUAS) equipped with various sensor payloads to perform object detection across the compromised airfield environment. Hyperspectral imaging has been specifically targeted as a promising sensor solution due to its enhanced discriminatory power in classifying materials. This study is focused on understanding how measurements of these small objects are affected by changes in parameters that govern operation of the drone-sensor system. Radiometric precision and spatial resolution are evaluated with respect to changes in flight speed, altitude, shutter speed, gain, and frames per second, in realistic field conditions. Within the ranges evaluated for each system parameter, the drone-sensor system presented spectrally and spatially resolves objects captured by just a few pixels with sufficient accuracy and precision for the RADA application.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Amirhossein Hassanzadeh, Jose N. G. M. Macalintal, David Messinger
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 1303109 (2024) https://doi.org/10.1117/12.3012402
Data fusion, which involves integrating data from multiple sources, is increasingly valuable across various fields due to its ability to enhance information quality, accuracy, and reliability. This process enables a more comprehensive understanding of complex phenomena by merging diverse datasets, providing insights that are otherwise unattainable. In the realm of remote sensing, where precise data acquisition is critical, fusion techniques have become indispensable, benefiting applications such as object detection, classification, and change detection. While much emphasis has been placed on spatial sharpening techniques in published studies, there remains a notable gap in establishing robust workflows for both lab-based and UAS-based remote sensing data fusion, particularly in the near-infrared (VNIR) and short-wave infrared (SWIR) regions. This study aims to investigate VNIR-SWIR fusion using data sourced from a medieval manuscript in a controlled laboratory environment and from UAS-based sensors in a real-world setting, addressing differences in system parameters and processing workflows. Despite challenges such as image registration issues, our analysis has yielded promising results, underscoring the importance of ongoing refinement in fusion methodologies to ensure comprehensive data interpretation and analysis across diverse datasets and environments.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Spectral Sensing for Space Situational Awareness: Joint Session with Conferences 13031 and 13062
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310A (2024) https://doi.org/10.1117/12.3013435
The proliferation of objects in space is a growing concern, and it is becoming crucial to have reliable information about these objects in order to be able to locate and identify them. The aim of this paper is to present SIRIUS (Simulateur InteRactif d’Images Ultra Spectrales) tool dedicated to optical modeling of resident space objects (RSO) in order to design or configure ground-based or space-based future observation systems. This tool presents an innovative feature as it generates hyperspectral highly spatially resolved images, at low computation time. Images are physically realistic thanks to the use of a global illumination method implemented through optimized rendering methods on GPU. The optical signature of a RSO is calculated in the visible range, with a possible extension to infrared. The radiative environment consists of the Sun and the Earth (ground, and possibly atmosphere and clouds), and the reflectivity of the RSO is described by Spectral Bidirectional Reflectance Distribution Function (sBRDF) for each material with physical modelling of geometry wrinkling for MLI materials. The orbit, orientation and mobile parts rotation (solar panels of satellite) are taken into account dynamically over time. The result of the simulation consists of a series of spectral images at the sensor position, for an observer located in space or in the Earth’s atmosphere. The optical effect of turbulence and the sensor may be modeled using a Point Spread Function (PSF). In this paper, we propose a validation of the SIRIUS tool based on a comparison of a simulated light curve obtained from ENVISAT images taken with the MeO telescope at the OCA (Côte d'Azur Observatory, France) during 2019, and an evaluation of the MLI wrinkling on the signature. The comparison between the simulated and experimental data of this uncontrolled satellite highlights the performance of this new predictive tool.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310B (2024) https://doi.org/10.1117/12.3014277
This work advances Space Situational Awareness (SSA) by analyzing ground-based hyper/multispectral images of Unresolved Resident Space Objects (URSO). Machine-learning models are constructed for satellite classification using unresolved spectral imagery. The study uses simulated data of observations of nine distinct satellites retrieved from the U.S. Space Force Unified Data Library (UDL). The dataset consists of unresolved hyperspectral imagery of satellites in different poses collected with a slitless spectrograph imager. The slitless spectrograph allows the collection of the spectral signature of the URSO with a single focal plane array in the 630 nm to 980 nm spectral range. In practice, the expectation is to have more unlabeled than labeled samples. Thus, the proposed classifier leverages the concept of Semi-Supervised machine learning. The network architecture leverages image reconstruction via a Convolutional AutoEncoder (CAE) and Multi-layer Perceptron (MLP) for classification. The architecture consists of three submodels: an Encoder, a Decoder (the two traditional components of a CAE), and a separate MLP. The Decoder and MLP use the Encoder’s low-dimensional representation of the samples as their input. The Encoder performs the task of creating the shared latent space via dimensionality reduction of the spectral data. The CAE (Encoder/Decoder) can be trained on labeled and unlabeled data, while the MLP requires labeled data for training. The newly developed classification models achieve a mean validation accuracy of 84% and a mean testing accuracy of 82%, utilizing 10-fold cross-validation. Additionally, the decoder achieves low image reconstruction error with a mean test error of 0.008, measured by Mean Squared Error, over the cross-validation folds. This network architecture demonstrates the ability to map slitless spectral data for accurate identification of RSO despite the challenges of using a limited-size and unbalanced dataset.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310C (2024) https://doi.org/10.1117/12.3012882
Diffusion models are a promising generative artificial intelligence (AI) technique for denoising and synthetic data generation. A de-noising diffusion model is applied to the atmospheric correction problem. In place of true noise, the model is trained based on predictions from the physics-based atmospheric radiative transfer tool, MODTRAN, to constrain the training environment. In this paper, we present results from a trained diffusion-based neural network model applied to hyperspectral image data and assess performance compared to conventional empirical atmospheric correction algorithms.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310D (2024) https://doi.org/10.1117/12.3013893
More than 65 million tons of plastics and organic materials from municipal solid waste (MSW) typically end up in landfills unless alternative routes can be found for their use. Given the high volume and low cost of these materials, they represent an attractive option to develop technologies for producing cost-competitive advanced biofuels from non-food biomass resources. These up-cycling technologies will require real-time composition analysis to operate properly and efficiently. A convolutional neural network (CNN) is being developed to classify components of MSW based on their visual and midinfrared (MIR) quantum cascade laser-based spectroscopy. We test this classifier by streaming labeled visual and MIR spectral images to simulate its operation in real-time on moving waste streams. These simulations allow us to explore how performance parameters like speed and accuracy are related to the required preprocessing of input the CNN and the structure of the network.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310E (2024) https://doi.org/10.1117/12.3013843
Identification of solids via infrared reflection spectroscopy requires a spectral library of all solids likely to be encountered. A confounding factor in populating such a spectral library is that the reflectance spectra of solids vary with their form, including particle size, film thickness, and substrate. To reduce the efforts of experimentally constructing such a library, an alternate strategy is to use the wavelength-dependent optical constants, n and k, of a solid to calculate a series of reflectance spectra corresponding to each scenario or morphology. Because most n/k measurements are best performed on mm-sized crystals, however, the challenge of determining the optical constants increases when a solid is only readily available as a powder, as is often the case. Some organic solids, such as caffeine, are both unavailable in large crystals and difficult to press into pellets. In this study, the infrared optical constants, or complex refractive indices, of caffeine were determined using three different methods: single-angle reflectance, infrared spectroscopic ellipsometry, and quantitative absorbance measurements of KBr pellets. The n and k values derived through each method were used to model the hyperspectral imaging reflectance spectrum of a caffeine film on a steel planchet. Over 1,110 – 870 cm-1, the single-angle reflectance-derived n and k had the best correlation with the experimental spectrum. These results suggest different organic solids may require different methods to determine the most accurate infrared complex refractive indices for synthetic spectral libraries.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310F (2024) https://doi.org/10.1117/12.3013830
The next generation of infrared spectroscopic solutions collect a massive amount of data that is realistically much too dense to be intuitively understood by a human. Thus, as a practical necessity, the user is generally interested in a smaller number of “latent” variables that aren’t directly observed. However, the usual method of considering a more manageable subset of the raw data throws away a great deal of collected information. The problem of distilling the latent variables and related uncertainties from the raw data is one of statistical inference. We adopt a Bayesian approach to better quantify the uncertainties in the latent variables. While our prior work has focused on exact inference methods such as Gibbs Sampling and Hamiltonian Monte-Carlo, we have begun exploring techniques for approximate inference, which allow such data analysis to proceed in quasi real-time.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Machine Learning Applications in Hyperspectral Imaging
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310H (2024) https://doi.org/10.1117/12.3013758
We seek to detect and classify chemical threats based on their infrared spectra. Specifically, we are interested in utilizing spectral signatures observed with standoff technologies that interrogate analyte micro-particles on relevant substrate surfaces such as glass, metal and plastics. In this work, we have applied six Machine Learning algorithms to classify analytes based on their infrared spectra. Two synthetic datasets were used, the first one containing 40 analytes and the second one containing 55 analytes. In both datasets the analytes were synthetically placed onto 9 substrates. The 40 analytes dataset contains 18,000 spectra, 450 for each analyte with mass loading varying from 1 to 50 μg/cm2. The 55 analytes dataset consists of 49,500 spectra, 900 for each analyte and mass loadings in the range 1 to 100 μg/cm2. Two of the algorithms used in this work are coming from the statistical field; k nearest neighbors (k-NN) and Logistic Regression. The Support Vector Machine algorithm was developed by the Machine Learning community. Multilayer Perceptrons (MLP) as well as Convolutional Neural Networks are considered Deep Learning Algorithms. In addition to that, we have considered the hybrid deep learning algorithm one dimensional CNN-LSTM. Our experimental results lead us to the conclusion that k-NN and logistic regression outperform deep learning algorithms for our synthetic data sets. However, after dimensionality reduction using PCA, the accuracy of k-NN decreases and the performance of deep learning algorithms improves. We also considered the effect of mass loadings and added noise on the performance of the classifiers.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310I (2024) https://doi.org/10.1117/12.3012800
The high spectral resolution afforded by Hyperspectral Imaging (HSI) sensors is poised to bring unprecedented advancements to signature characterization applications. Thus far, much of the research in the machine learning field devoted to HSI applications has focused on a few specific tasks like land-use/land-cover classification. In land classification tasks, spatial information is very important, and model architectures are often designed to leverage spatial contexts. However, it is unclear how well these spatially-tuned models will translate to tasks where spectral information is critical, like the detection and characterization of chemicals. In this work, we compare spectral models (inputs are 1D spectra) and spatial-spectral models (inputs are 3D cubes) in the context of predicting chemical concentration maps. We find that spatial-spectral models perform the best, though we find a wide range in performance across the different architectures tested. Additionally, we find that model performance is impacted by the availability of training data, particularly in scenarios where the training data doesn’t fully capture the true variance of real-world conditions. We find that data augmentation can help mitigate sparse coverage of observed parameter space (e.g., seasonal or geographic variability in ground cover), and present augmentation strategies that are tailored to hyperspectral data.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Noriaki Kono, Eleanor B. Byler, Charles W. Godfrey, Timothy J. Doster, Tegan H. Emerson
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310J https://doi.org/10.1117/12.3013487
Given the scale and complexity of forthcoming HSI data, producing labeled datasets at the scale required to improve state-of-the-art performance is impractical and prohibitively costly. Unsupervised pre-training algorithms have revolutionized deep learning for natural language processing and computer vision by tapping into vast troves of unlabeled data, but these advances have seen little adoption in the HSI domain. We present some early results from self-supervised pre-training for hyperspectral imagery using masked auto-encoders early and compare different pre-training approaches and masking techniques; specifically masking size, dimension (spatial, spectral, both), mask fraction, and mask coherence (spatially independent or consistent). We summarize our lessons learned and highlight the most promising approaches towards building a foundation model for hyperspectral data.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310K (2024) https://doi.org/10.1117/12.3013984
In support of the detection of explosives and threat chemicals by active infrared backscatter hyperspectral imaging, we are training algorithms to process and alert on possible threats. Surfaces are interrogated using infrared quantum cascade lasers (QCL) and the backscattered signal is collected using a cooled MCT focal plane array (FPA). The QCLs can tune across their full wavelength range, from 6 – 11 m, in less than one second. Full 128 X 128 pixel frames from the FPA are collected and compiled into a hyperspectral image (HSI) cube containing spectral and spatial information from the target. The HSI cubes are processed and the spectra from extracted pixel locations are then run through an algorithm to detect and identify traces of explosives. We train our algorithms on both synthetic and experimental data. In this presentation, we utilize machine learning algorithms to classify HSI cubes from a series of targets coupons fabricated on relevant substrates (glass, painted metal, plastics, cardboard). We explain how the algorithm training uses reference spectral measurements from our cart system as well as from a benchtop FTIR. The generation and utility of synthetic data is described regarding how we populate the algorithms’ spectral library more densely than would be possible using only measured experimental data. The performance of several ML algorithms is described.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310L (2024) https://doi.org/10.1117/12.3013935
Satellites are equipped with diverse sensors, capable of capturing detailed information across a multitude of wavelengths. The fusion of multispectral data is pivotal to amplify the visual representation of the area of interest. The improvement of information representation allows for enhanced processing, analysis, and other crucial tasks for numerous fields of study, including remote sensing, defense, and material characterization. Previous solutions often utilize traditional signal processing techniques, including principal component analysis (PCA), to accomplish data fusion. By performing fusion on a feature level, extracted information about the area of interest texture and boundaries are combined. The introduction of neural network techniques improved the reconstruction of data similar to the results obtained by conventional inference of humans. For example, the use of deep learning algorithms in conjunction with PCA allowed for refined reduction of redundancy and distortion of spectral data, in comparison to traditional methods alone. The introduction of the Vision Transformer (ViT) architecture, originally developed for two-dimensional image data, has revolutionized image processing tasks, vastly improving performance at the cost of a large quantity of trainable parameters. Recent experimentation has proven that optimizing ViT for efficiency allows for comparable or even superior performance while lessening the computational cost. The transition from 2D to 3D information via utilization of additional depth and spatial data has also led to superior results as the added information allows for better representation of terrain features, making it invaluable for satellite imagery analysis. Combining the principles of ViT and 3D information to process complex satellite data can result in more effective data fusion to achieve a superior level of data visualization of multispectral satellite imagery in an efficient manner.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310M https://doi.org/10.1117/12.3011190
A hazardous noxious substance (HNS) spill accident is one of the most devastating maritime disasters as it is accompanied by toxicity, fire, and explosions in the ocean. In this study, a ground HNS measurement experiment was conducted for artificially spilled HNS by using two hyperspectral cameras at VNIR and SWIR wavelengths. HNS images were obtained by pouring 1 L of toluene into an outdoor marine pool and observing it with a hyperspectral sensor installed at a height of approximately 12 m. The pure endmember spectra of toluene and seawater were extracted using principal component analysis and N-FINDR, and a Gaussian mixture model was applied to the toluene abundance fraction. Consequently, a toluene spill area of approximately 2.4317 m2 was detected according to the 36% criteria suitable for HNS detection. The HNS thickness estimation was based on a three-layer two-beam interference theory model. Considering the detection area and ground resolution, the amount of leaked toluene was estimated to be 0.9336L. This study is expected to contribute to the establishment of maritime HNS spill response strategies in the near future based on the novel hyperspectral HNS experiment.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
J. Duncan, R. Viger, T. Mayo, S. Ramsey, E. Michaelchuck Jr., S. G. Lambrakos
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310N (2024) https://doi.org/10.1117/12.3011635
This study describes case studies for inverse spectral analysis and parametric modeling of diffuse reflectance spectra for NIR-SWIR absorbing dyes. These case studies demonstrate the concept of applying inverse spectral analysis to diffuse reflectance, for estimation of absorbance functions, and parametric modeling for simulation of diffuse reflectance. Sufficient sensitivity of absorption spectra relative to inverse spectral analysis establishes that estimated absorbance functions can be used for parametric modeling of reflectance from dye formulations on substrates, e.g., fabrics.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310O (2024) https://doi.org/10.1117/12.3012711
Hyperspectral imaging is one of the prevailing tools for the analysis of material composition and feature identification in images. The spectral features contained within image pixels are studied to better understand a pixel’s material composition. However, a major limitation of using a single hyperspectral image sensor is that not all distinguishing spectral features of an image are captured within the conventional operating wavelength bands of a single hyperspectral sensor. While a straightforward solution of using a single hyperspectral sensor with larger wavelength ranges can address such an issue, this is often met with external limiting factors such as sensor availability or costs of operations. Therefore, in order to overcome the limitations of a single-sensor approach, there is motivation to integrate information from multiple sensors instead to create a more useful and accurate representation of a real-world scene. A particular challenge is that different sensors typically produce images of the same scene but at different spatial and spectral resolutions. In this study, we attempt to fuse VNIR and SWIR hyperspectral images and evaluate the fused images’ performance as applied to Cultural Heritage. Specifically, image fusion is performed on a medieval manuscript referred to as the Italian Leaf provided by the University of Durham.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310P https://doi.org/10.1117/12.3012904
In this study, we conducted field experiments aimed at acquiring spectra of Hazardous and Noxious Substances (HNS) and examining their variations. Spectral measurements were conducted at an outdoor pool situated in the CEDRE in Brest, France. Hyperspectral cameras equipped with 160 bands covering the range from 400 to 1,000 nm were employed to capture the spectra. To elucidate the spectral characteristics of HNS, we extracted the radiance values of HNS and conducted a quantitative analysis of the spectral patterns. While different behaviors were observed depending on the specific HNS under investigation, overall, the radiance values of HNS spectra at the peak wavelengths exhibited variations in response to wind conditions. For highly volatile HNS such as toluene and xylene, the radiance difference compared to the surrounding seawater increased under the influence of wind, but subsequently decreased beyond a certain wind speed. Conversely, the radiance difference of the condensate decreased as wind speed increased. This demonstrates the utility of quantitative analysis in enhancing our understanding.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310Q (2024) https://doi.org/10.1117/12.3013227
The accumulation of microplastics (MPs) in different environmental compartments represents a real emergency with dangerous effects on all ecosystems and human health. MPs analysis by the commonly adopted methods (i.e. FT-IR or Raman spectroscopy) is time-consuming, limiting the ability to monitor and mitigate plastic pollution. In this context, hyperspectral imaging (HSI) can be considered a promising identification tool, allowing the possibility to obtain rapid classification maps of MPs in different environmental matrices. In this work, an innovative application of HSI technology in the short-wave infrared range (SWIR: 1000-2500 nm) for rapid recognition and classification of MPs in real beach sand samples, coupled with machine learning approaches, is presented and discussed. MP samples were collected during a sampling campaign at Torre Guaceto beach (southern Italy), located along the Adriatic flank of the Apulia region, belonging to a natural protected area. Different spectral preprocessing strategies were tested on the acquired hyperspectral images in order to build a classification model capable of recognizing the complex mixture of materials that constitute MPs and beach sand matrices. The results of the study demonstrated as the proposed approach represents a powerful, fast and effective alternative to the most common adopted analytical methods for MP classification.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310R (2024) https://doi.org/10.1117/12.3014030
The utilization of hyperspectral image data has contributed to improved performance of machine learning tasks by providing spectrally rich information that other more common sensor data lacks. An issue that can arise when using hyperspectral imagery is that it can often be computationally burdensome to collect and process. This study seeks to investigate the incorporation of hyperspectral image data collected on a co-aligned VNIR-SWIR sensor for the purpose of hyperspectral image classification. In which, the evaluation is focused on investigating the distinct effects pertaining to the VNIR data, to the SWIR data, and to the combination of the two data types with regards to hyperspectral image classification performance on vehicles. The experiments were run on data collected by the US Army Corps of Engineers Research and Development Center.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310S (2024) https://doi.org/10.1117/12.3014862
An innovative approach based on hyperspectral imaging (HSI) was developed to monitor commercial starch-based (MaterBi®) disposable bioplastic behavior during anaerobic degradation. Mater-Bi® (MB) tableware items were selected among the ones available in supermarkets and compliant with the EN 13432 standard (EN 13432:2008). The MB items were manually cut in fragments with size ranging from 0.5 to 2 cm, removing the edges and the bottom to ensure test material homogeneity in terms of thickness. The anaerobic sludge was collected at a full-scale mesophilic anaerobic digestion plant treating a mixture of organic residues from food industries and was used as inoculum. Hyperspectral images of the samples composed of MB fragments dispersed in the anaerobic sludge were acquired in the short-wave infrared range (SWIR: 1000-2500 nm). A chemometric approach was then developed to analyze HSI data. In more detail, Principal Component Analysis (PCA) was applied for data exploration, followed by the implementation of a classification model based on Partial Least Square-Discriminant Analysis (PLS-DA) able to identify MB in the sludge. The achieved results are very promising, especially with reference to the possibility to adopt a fast strategy to monitor the behavior of MB during the anaerobic biodegradation process.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310T https://doi.org/10.1117/12.3014333
Remotely monitoring coastal ecosystems is critically enabled by hyperspectral imaging (HSI). HSI is corrupted by the water column in the scene, imparting a fingerprint on the imagery that is characteristic of the scene. Removing these effects from the scene requires (i) radiative transfer modeling and simulation, or approximations thereof, and (ii) knowledge of the sub-surface topology, i.e. bathymetry. This work explores using a novel physics-informed machine learning (ML) framework for joint estimation of bathymetry and corrected spectra. We formulate the problem of transmission of spectra through a medium as a tunable Ordinary Differential Equation (ODE). Here, the learned ODE is a surrogate model for transmission of radiation while the integration path length provides a measure of column depth; i.e. bathymetry. In this work, we show that inverting this ODE-based relationship provides reasonable estimates for bathymetry. We demonstrate a proof-of-concept of this inversion using a HSI scene of Enrique Reef on the southwest coast of Puerto Rico.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310U (2024) https://doi.org/10.1117/12.3013807
In this article the results of the disruption module in the hyperspectral image field are analyzed and present itself. The aim is to see the major concerns, techniques and potential areas that can be researched as topics in the bibliometric method. Highly specified hyper-spectral spectroscopy that can simultaneously analyze the spectral data beyond the visual spectrum down even to the nanoscale. It can be used for various reasons, including environmental management, agriculture, and mineralogy. Unlike small data systems, where patterns are simple to identify, in large scale systems, where there is huge data quantity to study, there should be a wide variety of complexity of algorithms to use in reliable anomaly detection. This research bases its methodology on a two-layer mechanism which is simply to conduct very thorough literature review and deep bibliometric analysis to showcase the innovations in the field of anomaly detection of hyperspectral radar. The role of the states of estimation and the parameters, trained and objective, filtered, control rules, and fuzzy logic are some aspects, among the many, that is coming to the fore during the anomaly detection process. This paper aims to carry out a detailed analysis to mainly focus on the technical and methodological advancements that have reshaped the research area. It also shows that priority should be given to this aspect and that anime detection is the most challenging part of hyperspectral. What is more, this process gives numerous valuable signals about potential routes for future research. The results point to the dominant feature of the developed strategy and analysis which was mostly based on the special of multi-dimensional and transdisciplinary thinking
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Proceedings Volume Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXX, 130310V (2024) https://doi.org/10.1117/12.3017671
The article presents methodological approaches to assessing the expected socio-economic effect of reducing the resource potential of the tourism business in conditions of increased possibility of natural and artificial emergencies. The research aims to use Earth remote sensing technologies in recreational areas with subsequent socio-economic analysis to reduce the resource potential in the tourism business in the context of the deteriorating ecology of the territories. The work uses regression analysis, methods for deciphering satellite images using various techniques, mathematical modelling and the formation of forecasts using regression models. A general methodology is proposed for constructing a geo-ecological model using information generated from data from space, Earth remote sensing technologies, and monitoring territories for waste sites with a socio-economic justification for tourism business in recreational areas. A mathematical model for the formation of tourist flows using remote sensing technologies has been developed, as well as a methodology for forecasting tourist flows using PLS and PLS-LM methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.