This PDF file contains the front matter associated with SPIE Proceedings Volume 12094, including the Title Page, Copyright information, Table of Contents, and Conference Committee listings.

Hyperspectral imaging has over the last thirty years developed into a power analytical tool for the determination of chemical and other properties. As a result, there has been strong development in both the design of spectral cameras and in the applications for which they are used. This has led to a diversity in the way fundamental instrument performances are characterized, reported, and understood. As a result, this makes it difficult to compare instruments for applicationspecific needs, or for commercial market needs. In 2018, the IEEE P4001 group was formed to facilitate the development of a standard to unify the use of terminology, spectral camera characterization methods, and the meta-data structures that are needed to represent spectral camera performance. This talk provides an update on the work to date, and the significant progress made towards the first draft of the standard.

The IEEE P4001 group is developing a new standard for hyperspectral imaging. A central part of the work is to define a standardized way to specify hyperspectral cameras. P4001 aims to define a complete description of the camera as a "black box", i.e. without referencing camera internals such as f-number. The aim is to use characteristics that are welldefined physical quantities, and which capture all important aspects of performance, in some cases with a tradeoff to avoid the need for excessive testing equipment or effort. The characteristics apply to the most widely used camera architectures based on dispersive spectrometers or tunable filters.

Hyperspectral images are ultimately analyzed by users for specific applications. Metadata supplied with the images are critical to enable the user to extract full information from the data. An important part of IEEE P4001 Hyperspectral Standard is therefore to define metadata, which will be physics-informed and related to the camera characteristics. With metadata such defined, it will for example be possible for a user to estimate signaldependent noise in the images. Different applications will require different levels of metadata. P4001 therefore defines notional use cases for machine vision, laboratory, and Geospatial Imaging, in order of increasing need for metadata. P4001 will define required and optional metadata for these use cases.

Data standards enable low-cost integration and interoperability for data hosted by an enterprise managing multiple data sources, yet standards must change to keep pace with technological advances. Hyperspectral imagery (HSI) systems have evolved beyond the 2011 tactical, i.e., airborne, HSI standard for systems transmitting imagery using the National Imagery Transmission Format (NITF) and hosted within the National System for Geospatial Intelligence (NSG). The Research Directorate of the National Geospatial-Intelligence Agency (NGA) created the Spectral NITF Implementation Profile (SNIP) to realize HSI systems’ advanced capabilities, incorporating improvements in remote sensing made over the past decade. Ratified by U.S. government standards committees, the SNIP is a mandatory Department of Defense and Intelligence Community standard that applies to any spectroradiometric system regardless of the number of bands. The SNIP supports spectroradiometric exploitation, geopositioning and mensuration, search and discovery, and dissemination controls. SNIP improvements include support for air, space, and ground-based platforms; new geopositioning extensions that provide more complete and accurate results; and per-pixel metadata that include spectral smile, noise equivalent spectral radiance, and keystone. SNIP reference implementation products (RIPs), that is, datasets encoded to the SNIP standard, are derived from a U.S. National Aeronautics and Space Administration (NASA) Hyperion dataset. The publicly released RIPs support interoperability tests and algorithm development. During development, the RIPs test dataset design, sometimes revealing issues that can be resolved with a revised design. This paper describes the SNIP, including development premises, design, collaboration efforts, history, and plans. It also describes the RIPs and includes a short overview of the NITF standard.

Physical Sciences Inc. has developed an ultra-compact shortwave infrared (SWIR) staring mode hyperspectral imaging (HSI) sensor with an additional visible full motion video (FMV) capability. The innovative HSI design implements a programmable micro-electromechanical system entrance slit that breaks the interdependence between vehicle speed, frame rate, and spatial resolution of conventional push-broom systems and enables staring-mode operation without cooperative motion of the host vehicle or aircraft. The FMV and HSI components fit within 1000 cm³, weigh a total of 2.1 lbs., and draw 15 W of power. The sensor mechanical design is compatible with gimbal-based deployment allowing for easy integration into ground vehicles or aircrafts. The FMV is capable of achieving NIRS-6 imagery over a 6°×6° field-of-view (FOV) at a 1500 ft. standoff. The SWIR HSI covers a spectral range of 900-1605 nm with a 15 nm spectral resolution, and interrogates a 5°×5° FOV per 1.6 s with a 2.18 mrad instantaneous FOV (1 m ground sample distance at 1500 ft.). A series of outdoor tests at standoffs up to 300 ft. have been conducted that demonstrate the payload’s capability to acquire HSI information. The payload has direct utility towards diverse remote sensing applications such as vegetation monitoring, geological mapping, surveillance, etc. The data product utility is demonstrated through the spectral identification of materials (e.g. foam and cloth) placed in the sensor’s FOV.

This paper presents the latest Imec snapshot hyperspectral imager based on either 3x3 or 4x4 mosaic filter patterning on an industry-ready SWIR detector. The mosaic patterns are implemented by means of high-transmission Fabry-Pérot interferometers processed using thin-film technology. Our snapshot hyperspectral imager offers a spatial resolution of 640x512 pixels down sampled according to the mosaic pattern to acquire data in 9 (3x3), or 16 spectral bands (4x4) in the 1100-1650nm range. To achieve imaging at the native spatial resolution of the sensor, super resolution methods are available post-acquisition. Moreover, this compact USB-3 camera of 65x65x130 cm and 260 gr (without lens) reaches an acquisition speed of up to 120 hyperspectral cubes/second, and it is therefore suitable for high-speed inspection such as a conveyor belt. Its potential on the sorting and recycling industry is showcased for three applications: plastic recycling application, textile sorting and discrimination of a mixed plastic and textile materials. The first application focuses on sorting of plastic (HDPE, PET, PP) and paper (cardboard, paper, newsprint, tetra Pak). The second application focuses on discrimination of textiles such as cotton, PET, wool, silk, polyamide, and blends of these ones. The last application aims at discriminating a new sample set of the previous categories of plastic and textiles. A pixel-accuracy over 90% is obtained for each sample set with close to 100% object-level discrimination. This shows the high potential of SWIR range snapshot hyperspectral imagers for sorting/recycling and, by extension, for many other applications.

Telops has a 20-year history in the design, construction, and deployment of thermal infrared hyperspectral imaging systems through the legacy Hyper-Cam line. Advances in critical subsystem technologies has allowed Telops to develop the next-generation of hyperspectral imaging systems with significant reductions in size, weight, and power requirements while maintaining imaging and data quality performance. This reduction in SWaP requirements yields a significant increase in deployment flexibility, allowing for increased capability for collecting actionable hyperspectral data of remote or difficult-to-access targets. Also taking benefit of the most recent data processing capabilities from modern electronics and computer systems, the real-time data analysis has enabled unprecedented ease of use and conviviality without compromise to performance. This presentation will serve as an overview of the system architecture and analysis capabilities of three next-generation thermal infrared hyperspectral imaging products. These platforms include a tripod-based system for ground measurements; an airborne platform designed for small, fixed-wing aircraft; and a small-footprint system designed for deployment on a quadcopter or other small UAV.

Starting in 2023, the Carbon Mapper public-private partnership will launch two imaging spectrometers into low earth orbit as the first demonstration satellites for a larger, emerging constellation. This mission is a critical collaboration between several partners including Planet, Carbon Mapper, Arizona State University, NASA’s Jet Propulsion Laboratory, the University of Arizona, the High Tide Foundation, California Air Resources Board, and the Rocky Mountain Institute. This hyperspectral constellation will complement Planet’s existing high-spatial and high-temporal mission lines and increase the ability to measure and monitor the impacts of climate change on our planet and tackle dynamic, wide-ranging and complex challenges such as sustainability. Each satellite is equipped with a 400 - 2500 nm hyperspectral imaging system capable of addressing a wide range of applications. The core mission for the Carbon Mapper Mission is to monitor climate risks (methane, CO2) but it has capacity to collect data for other sectors such as Defense, Intelligence, Agriculture, Mining, and others. The Carbon Mapper Mission is a tasked system and is designed to be responsive to dynamic events where analysis in a matter of days or hours may be important. In this paper, we provide an overview of the anticipated technical capabilities of the system and discuss applications for the Defense and Intelligence communities. We will also outline how the Carbon Mapper Mission can work in conjunction with the rest of the Planet constellations to enable unique fusion products.

We are developing machine learning algorithms to identify chemicals of interest by their diffuse infrared (IR) reflectance signatures. For capturing the signatures themselves, we are developing a cart-based mobile system for the detection of trace explosives on surfaces by active infrared (IR) backscatter hyperspectral imaging (HSI). We refer to this technology as Infrared Backscatter Imaging Spectroscopy (IBIS). A wavelength tunable multi-chip infrared quantum cascade laser (QCL) is used to interrogate a surface while an MCT focal plane array (FPA) collects backscattered images to comprise a hyperspectral image (HSI) cube. The HSI cube is processed and the extracted spectral information is fed into an algorithm to detect and identify chemical traces. The machine learning algorithm utilizes a convolutional neural network (CNN) that has been trained on synthetic diffuse reflectance spectra. In this manuscript, we utilize a CNN to identify chemicals within an IBIS hypercube. We demonstrate a form of active chemical imaging where the CNN identifies a chemical within each pixel of an IBIS hypercube. Chemical imaging capability goes beyond point detection and identification to indicate where each chemical is within the field of view, as well as identifying multiple target chemicals simultaneously.

Machine learning approaches, such as deep neural networks, have shown recent success for target detection and identification problems in hyperspectral imagery. However, when deployed “in the wild,” there are no guarantees about the behavior of these black box algorithms when encountering new materials or environmental conditions that were not part of the training data. In addition, neural networks typically lack properties of linear identification methods in that their predictions tend to select a single class with high confidence even when there are multiple classes that could match a given input spectrum. To provide estimates of confidence in neural network predictions (i.e., target identifications) and to produce indicators of uncertainty, we apply stateof-the-art uncertainty quantification techniques to neural networks trained on hyperspectral data. Specifically, we assess recently proposed methods from the machine learning community including Monte Carlo dropout, ensembles of neural networks, and variational Bayesian neural networks. We report not only the accuracy of the resulting model-averaged networks on in-distribution data, but also the usefulness of uncertainty metrics on noisy or out-of-distribution data. We also compare ensemble neural network target identification results to a linear method on airborne long-wave infrared (LWIR) hyperspectral data with real targets. Finally, we offer some guidelines for applying these methods to hyperspectral target detection/identification problems.

A platform for building sensor specific machine learning detection algorithms has been developed to classify spectroscopic data. The algorithms are focused on long wave infrared reflectance (LWIR) and Raman spectroscopies. The classification algorithm is based on a one dimensional (1D) convolutional neural network (CNN) architecture. Training data is generated using an appropriate signal model that is combined with sensor specific characteristics such as spectral range, spectral resolution, and noise. Within this paper, the performance of trained CNNs for both LWIR and Raman sensor systems has been evaluated. The evaluation uses both real and synthetic data to benchmark the performance in terms of the discriminant signal. The evaluation data consists of various chemical representations and varied noise levels. The performance of the 1D CNN approach has demonstrated high classification accuracies on data with low discriminant signals. Specifically, the CNNs have demonstrated a classification accuracy <90% for infrared reflectance data down to a wavelength averaged discriminant SNR<1. For Raman systems, we have demonstrated classification accuracies <90% for data with a peak discriminant SNR of approximately 6.

Effective space domain awareness (SDA) requires accurate positions and identities of artificial satellites. These measurements–critical to effective decision making in the high risk on orbit environment–are daunting in the deep space geosynchronous (GEO) regime. Here, distance precludes collection of spatially resolved measurements from ground-based telescopes. Neural networks designed for deep space object detection and spectroscopic positive identification have been shown to be effective tools for these mission critical SDA measurements. In this work we demonstrate the potential of slitless field spectroscopy to provide simultaneous object detection and identification of on orbit assets at GEO. Slitless spectrographs expose the reflection physics needed for spectroscopic positive identification without destroying the spatial information used for object detection. Such systems are compact and hardened in comparison to classic spectrographs, and may be deployed to small telescopes. In this work we present a GPU-accelerated simulation environment for the production of realistic synthetic imagery to support generation of large datasets for deep learning. We establish a baseline for simultaneous detection and identification performance by training convolutional neural networks on synthetic datasets created with this tool. This work reduces risk for initial technology development and dataset collection, and provides constraints to the design and development of slitless spectrograph systems for space domain awareness.

Tracking and classifying objects in the space domain is important for the military due to increased concern of congestion from commercial and national interests. This research use a 100-lines-per-millimeter diffraction grating to perform slitless spectroscopy across a network of telescopes to collect satellite spectral signatures. The pixel-to-wavelength relationship for the diffraction gratings are characterized using calibration stars with known spectral features. Once calibrated, the United States Air Force Academy (USAFA) telescopes are used to observe manmade satellites with the goal to characterize them based on their reflected spectra.

In this paper, performance of state-of-the-art partially supervised detection algorithms are compared. It can be difficult to characterize the performance of detection algorithms using field data, especially at the subpixel level, due to limited ground truth. Fortunately, the SpecTIR Hyperspectral Airborne Experiment (SHARE) 2012 contains multiple sets of targets for testing detection algorithms with excellent ground truth. In this paper we utilize field spectra acquired for six targets to evaluate the performance of multiple detection algorithms. Each method is initialized with a single field spectra target signature, and detection performance of each method is separately assessed for each of the six targets. Detailed evaluation of these detection methods on the SHARE 2012 hyperspectral data is provided.

Hyperspectral imagers allow for the identification of materials of interest (MOI) in a scene using spectroscopic analysis. Material identification is made more difficult when the MOI fills a fraction of the pixel. The resulting pixel spectrum is a linear mixture of the MOI and its surrounding background with weights/abundances based on the fraction of each material within the pixel. Some identification methods utilize pixel unmixing to match a library spectrum to the suspected MOI. The resulting quality of fit depends on how closely the library spectrum matches that of the MOI. Further analysis is accomplished by removing the background portion, and comparing the normalized MOI portion to the library spectrum. When the library spectrum corresponds to the MOI, the two spectra should visually match, with the exception of noise. Often, spectral smoothing is required to improve the match as the SNR of the MOI portion decreases with its abundance. We propose a spline-based smoothing method to reduce noise error greatly while maintaining the fine spectral features used to distinguish between spectrally similar materials. The main practical difficulty of spline smoothing lies in setting the parameter which determines the amount of smoothing. We show that automatically setting the smoothing parameter, such that the roughness of the smooth MOI portion matches the library spectrum, prevents over smoothing and improves identification.

Mucilage also called sea snot or sea saliva is a collection of mucus-like organic matter found in the sea. Although not harmful in the beginning, when mucilage increases over time, it covers the sea creatures and forms thick layers in the sea. Its existence and long duration change the oxygen balance in the seas, reduce biodiversity, fisheries, and tourism. Since April 2021, mucilage has emerged as both an ecological and economical problem in Turkey, spreading over an area of kilometers, clogging the fishing nets, causing problems in marine vessels, and disrupting the industry. These findings indicate that mucilage monitoring, early detection, and intervention before the economic and ecological damages grow out of proportion is quite necessary. Through the analysis of satellite data, it is possible to observe the existence of mucilage as thin, extended layers of white substance. Therefore, in this work, we analyze the Sentinel-2 multispectral data and PRISMA hyperspectral data to detect the mucilage in the early stages through the use of both traditional as well as deep learning algorithms. Our results indicate that it is possible to detect mucilage from satellite data with high accuracy, saving time and money for the cleaning efforts.

In this paper we derive two algorithms for estimating concentrations of a known chemical compound from compressed measurements of a hyperspectral image (HSI). It is assumed that each resolved pixel in a scene contains a chemical of known spectral signature, at an unknown concentration. The problem is to estimate the concentration directly from the compressed measurements. Estimated concentrations are either displayed or used as detection scores in a threshold test for presence or absence of chemical. In the first algorithm we use matched filtering and ℓ₁ regularization to extract an image of concentrations, directly from compressed data. In the second we model the image of concentrations in a fixed-resolution subspace of the 2D Haar wavelet domain, estimate its parameters in this space, and reconstruct the image of concentrations at a macro-pixel resolution. We evaluate our algorithms by applying them to several long-wave infrared (LWIR) HSI data sets, either synthetically generated or recorded by Physical Sciences Inc. Synthetically-generated data is compressed with a mathematically-defined linear compressor; real HSI data is compressed with PSI’s Digital Micromirror Device (DMD), which is a physical implementation of a mathematically-defined compressor; Fabry-Perot data is raw HSI data recorded by PSI, which is then compressed with a mathematically-defined compressor. We demonstrate for these data sets that estimating concentrations through matched filtering and ℓ₁ inversion of compressed measurements yields detection performance that is as good as previously proposed methods that first reconstruct a hyperspectral data cube from compressed data, and then estimate or detect chemical concentrations. The proposed methods save on memory and computation. We demonstrate that detection performance is maintained when resolving concentration maps at a lower resolution, so long as the resolution is not too low.

Leakage in pressurized water nuclear reactors (PWRs) poses potentially serious health and safety threats to workers, residents, and the public at large. Early detection and prompt reparative actions of such breaches are critical to both maintaining a secure reliable power supply and ensuring the public’s confidence in the safe operation of commercial nuclear power generation. Currently many components are inspected using visual examinations performed directly by examination personnel or by remote methods using cameras. When indications of leakage are determined to be relevant, additional methods such as chemistry and isotopic analysis are needed for characterization and source determination. This laboratory-based analysis is a burden on resources that may be able to be avoided by the deployment of enhanced imaging methods. Shortwave infrared (SWIR) hyperspectral imaging offers a potential means to characterize chemical species deposits, determine sources and avoid the need for burdensome substance sampling and laboratory analysis. This technique has shown excellent promise for identifying various chemical species in many industrial applications that may otherwise not be differentiable by conventional imaging or by eye. Recent developments in Fabry-Perot interferometerbased hyperspectral imaging have enabled near real-time image capture and classification, thus further expanding the utility of the technology and automating it for use in hazardous environments when paired with robotic platforms such as magnetic crawlers.

An environmentally hardened compressive sensing hyperspectral imager (CS-HSI) operating in the long wave infrared (LWIR) has been developed for low-cost, standoff, wide area early warning of chemical vapor plumes. The CS-HSI employs a single-pixel architecture achieving an order of magnitude cost reduction relative to conventional HSI systems and a favorable pixel fill factor for standoff chemical plume imaging. A low-cost digital micromirror device modified for use in the LWIR is used to spatially encode the image of the scene; a Fabry-Perot tunable filter in conjunction with a single element mercury cadmium telluride photo-detector is used to spectrally resolve the spatially compressed data. A CS processing module reconstructs the spatially compressed spectral data, where both the measurement and sparsity basis functions are tailored to the CS-HSI hardware and chemical plume imaging. An automated target recognition algorithm is applied to the reconstructed hyperspectral data employing a variant of the adaptive cosine estimator for detection of chemical plumes in cluttered and dynamic backgrounds. The approach also offers the capability to generate detection products in compressed space with no CS reconstruction. This detection in transform space can be performed with a computationally lighter minimum variance distortionless response algorithm, resulting in a bandwidth advantage that supports efficient search and confirm modes of operation.

Accurate human perceived color cannot be obtained from most multispectral imagery (MSI) sensors without postprocessing, as sensor spectral response is often not in parity with the spectral sensitivities of the human visual system. In this experiment, 30 WorldView-2 and WorldView-3, 8-band visible-through-near-infrared (VNIR) images were collected over a color array consisting of 24 16-ft × 16-ft plywood targets painted to match the Macbeth ColorChecker scheme. The images were atmospherically compensated using the Empirical Line Method (ELM), and 6-band reflectance spectra covering the visible wavelengths were harvested from the image for each color target. These reflectance spectra were reconstructed to high-resolution spectra using spectral basis functions derived from a principal component analysis (PCA) performed on a set of high-resolution spectra measured from a calibrated color standard. Initial reconstruction performance was evaluated by comparing reconstructed spectra to high-resolution ground truth spectra using the Root Mean Square Error (RMSE). Reconstructed spectra were then converted to CIELAB (L*a*b*) values using the D65 illuminant, and CIE 1931 2° Standard Observer. Reconstructed L*a*b* values were compared to ground truth L*a*b* values using the CIEDE2000 (ΔΕ*₀₀) color difference metric, and colorimetric accuracy of the process was evaluated. It was found that this method derived absolute colors with a median accuracy of ΔΕ*₀₀ < 6 across all color targets and images.

The nearest-neighbor diffusion-based algorithm (NNDiffuse) has seen great success in multispectral pansharpening. Here, we extend the capabilities of NNDiffuse to perform image fusion of high-res multispectral and low-res hyperspectral images (HRMSI+LRHSI fusion). Unlike learning-based frameworks which are computationally expensive and require extensive optimization and/or training time, NNDiffuse is fast, radiometrically accurate (introduces less spectral distortions compared to other state-of-the-art methods), and requires no training data. We introduce the utility of NNDiffuse in hyperspectral-panchromatic and hyperspectral-multispectral sharpening and look at workflows dealing with low-res HSI bands that fall outside the HRMSI spectral response functions. Sharpened image quality is assessed using image-wide metrics. Sharpening performance is also measured on the fused images in terms of their utility in pixel classification and ACE target-background separability. NNDiffuse and DL- and non DL-based methods are presented using hyperspectral imagery from SHARE2012, AVIRIS-NextGen, Hyperspec-VNIR-C, and ROSIS sensors.

Effective material identification with hyperspectral thermal infrared imaging requires an accurate estimate of the material emissivity within a pixel. When atmospheric effects are accurately known, the measured at-sensor radiance can be converted to ground radiance. Emissivity retrieval from the ground radiance is an undetermined problem due to the additional unknown temperature of the material. Hyperspectral temperature emissivity separation (TES) retrieval algorithms assume the emissivity spectra for solids are smooth and that the determined emissivity will be smoothest when the correct temperature is found. Smoothness-based TES algorithms typically quantify roughness by what a smoothing technique removes when applied to an emissivity estimate with temperature. We propose measuring the roughness of an emissivity estimate directly using splines. Splines can be used to fit a continuously differentiable function to the emissivity estimates at all temperatures exactly. Roughness can then be calculated as the integral of the second derivative squared of the spline. Where smoothing-based TES algorithms typically have difficulty on rough emissivity signatures due to the over-smoothing of true emissivity features, splines show resistance to rough signatures since they preserve the emissivity features. In this paper TES using splines will be described along with a comparative performance evaluation to the standard smoothingbased algorithm. Ground radiance test spectra are simulated with smooth and rough emissivity signatures. An improvement is shown using TES with splines for rough emissivity signatures.

Several different algorithms have been suggested for matched target detection for solid and gas point targets: theoretically, gas targets lead to an additive point target detection algorithm while solid targets lead to a substitutive point target detection algorithm. In this paper, we consider deviations that may occur in real life non-stationary backgrounds that would complicate the choice of algorithm. In particular, we have identified three possible weaknesses to the substitutive algorithm. First, incorrect estimates of the background for a minority of pixels can lead to considerably worse results in the substitutive algorithm compared to the additive algorithm. Second, an accurate knowledge of the amplitude (or power) of the target is needed for the substitutive algorithm compared to the additive algorithm. Thirdly, due to adjacency effects, it is not obvious which model to use in certain cases. For this latter case, we consider a hybrid model originally proposed by A. Ziemann et al. [1]. This model, with an adaptive parameter estimate of the form of the detection to use, appears to be a reasonable substitute for the two aforementioned algorithms.

An accurate statistical model for detecting sub-pixel hyperspectral targets requires the convolution of probability functions for all possible mixture ratios. That computational difficulty has been avoided in the past by accepting one or more compromises to natural probability models for backgrounds and targets. This paper meets the convolution problem headon, finding solutions that represent good approximations for two conventional methods of composite hypothesis testing. The methodology produces a bonus, a generalization of the most popular method of hyperspectral background modeling.

Transformers are becoming the state-of-the-art in multiple Computer vision (CV) and Natural language processing (NLP) tasks. As for hyperspectral target detection, a Transformer architecture named SpectralFormer2 has been developed and demonstrated improved performance over the previously state-of-the-art Convolutional neural network (CNN) architecture on widely studied classification tasks. The SpectralFormer was adapted from a CV architecture, the Vision Transformer (ViT). Concurrently, still in CV, a hierarchical and multi-scale version of the ViT, named Shifted windows (Swin) Transformer, is gaining momentum and is already the stateof-the-art on multiple tasks. In this paper, we adapt the Swin Transformer for hyperspectral classification and rare sub-pixel target detection. We apply this new architecture to commonly studied classification benchmarks public datasets such as Pavia University and Centre datasets, and to a new, large-scale airborne sub-pixel target detection dataset we developed. This new dataset is composed of over 100 M pixels taken over three collect days at three locations thousands of kilometers apart and in different climates. The proposed model reaches competitive performance on all of those tasks while reducing memory usage by 63 %.

It is often difficult to determine the precise characteristics that a hyperspectral sensor must have to meet mission objectives, but this information is crucial for designing a successful mission. We have developed a tool which emulates data from hyperspectral sensors by degrading other hyperspectral imagery. The tool performs target detection on the emulated imagery and assesses the utility of the modelled sensor based on detection performance. We will present details of how the tool functions; validation of the tool against airborne hyperspectral data; and extrapolations showing the estimated performance of different sensor configurations outside the validation envelope.

We intend to realize a non-invasive blood glucose sensor system based on long-wave infrared (LWIR) passive spectral imaging. We have developed an imaging-type two-dimensional Fourier transform spectrometer equipped with a microbolometer. Fourier transform spectroscopy is highly sensitive because of its high light-use efficiency. Our spectrometer is currently the only instrument capable of performing passive spectroscopic measurements at body temperature. The LWIR band is called the fingerprint region and is known to be highly suitable for component identification applications. In addition, LWIR radiation is emitted by the human body at room temperature, as can be observed in thermography. In active spectroscopy, light is used to illuminate the sample under test, and the amount of light absorbed as a result of molecular vibrations is measured. Therefore, in water-containing samples such as human skin, only the spectral characteristics near the skin surface can be observed because the absorption of the light by water is very strong. However, passive spectroscopy does not require illumination and thus can measure the amount of emitted light that arises from molecular vibrations. We therefore propose that it is possible to determine a subject’s blood glucose levels by measuring the radiation from the deep skin layer in which the blood vessels are located. In this report, we present passive spectroscopic measurements of a human forearm and report successful detection of the characteristic emission peak of glucose.

Remote assessment of physiological parameters has enabled patient diagnostics without the need for a medical professional to become exposed to potential communicable diseases. In particular, early detection of oxygen saturation, abnormal body temperature, heart rate, and/or blood pressure could affect treatment protocols. The modeling effort in this work uses an adding-doubling radiative transfer model of a seven-layer human skin structure to describe absorption and reflection of incident light within each layer. The model was validated using both abiotic and biotic systems to understand light interactions associated with surfaces consisting of complex topography as well as multiple illumination sources. Using literature-based property values for human skin thickness, absorption, and scattering, an average deviation of 7.7% between model prediction and experimental reflectivity was observed in the wavelength range of 500-1000 nm.

Graph Convolutional Network (GCN) is a graph-based learning framework that goes beyond the limitations of regular grid-sampled convolution operated by traditional CNN. By incorporating graph topological structure to the network, GCN is capable of encoding and aggregating features from neighboring nodes and exploiting the data nature in a non-linear way. For HSI data of cultural heritage artifacts, the lack of ground-truth labels imposes extra difficulty on the hyperspectral image classification task using a learning framework. For differentiating subtle spectral differences between varied pigments, we develop a two-layer semi-supervised GCN model for pigment classification, which can be directly applied to an adjacent region of interest (ROI) for automatically clustering unseen pixels without ground-truth labels. We first explore graph construction approaches in depth, and focus on two advanced graph construction strategies: Density-weighted kNN and Natural Nearest Neighbor (NNN). By assigning adaptive number of neighbors based on local data density, a local-scale adaptive graph structure can be built and infused into the GCN model. The experiments on the Selden Map and the Gough Map prove the validity of our method. Under semi-supervised learning framework, SAD+DWkNN GCN model can achieve near 90% classification accuracy for each class based on less than 30% labeled training samples. Compared to the ground-truth labels produced by the Graph Modularity algorithm, the GCN classification map matches even better to the actual spectral characteristics of each material via spectral analysis. The learning and generalization ability of the proposed GCN model shows a promising prospect for applying Graph Convolutional Network for a large-scale HSI clustering task. This research can alleviate the dilemma of deficient labeled samples and aid historians for analyzing varied pigments composition in a more intelligent manner.

In the petrochemical industry process water contains solid particles and oil droplets that need to be removed by the water treatment process before the water can safely be returned to the environment. The development of a hyperspectral imaging system for the detection and analysis of solid particles and the development of a fluorescence imaging system for the detection of oil droplets in in the process water of the petrochemical industry are presented. The particle and droplet size range for this application is in the range from 5 μm to 150 μm. Currently used particle analyzers use monochrome cameras with backlight illumination to image the particles in the process water. This results in a monochrome image which is used to derive shape and size of the particles and droplets. The shape information is used to differentiate between solid particles and oil droplets. The solid particles appear black in the images and therefore a more detailed analysis of their material is not possible. The high spectral resolution of the presented hyperspectral imaging system allows for an analysis of the particle materials. The fluorescence imaging system is used for a reliable detection of the oil droplets. This information is vital for the process water treatment in the oil production. The laboratory demonstrator and the initial results of measurements on solid particles and oil droplets in aqueous suspension are presented.

Development of portable Raman instrumentation over the past two decades has expanded in-field application of the method in areas including forensics, minerology, and plant pathology. Recently, Raman spectroscopy has attracted the interest of researchers from these fields because of its minimal interference from water and significant reduction of equipment costs. In this paper we demonstrate a hyperspectral Raman system capable of identifying minerals using very short integration times suitable for fast scanning of mining cores. Hyperspectral scans showed strong local Raman features of the known minerals in the core samples measured.

A numerical-analytical model and simulations are described concerning diffuse reflectance for surface-distributed material particles on substrates. The model combines an analytical formulation of Mie-scattering theory and a numerical procedure based on Kramers-Kronig analysis. The results of simulations using this model are compared with experimental measurements of diffuse reflectance for material particles distributed on a glass surface. The purpose of these comparisons is estimating the influence of background contributions to spectral features due to resonant scattering from finite-size particles. Evaluating the sensitivity of spectrum-feature extraction methodologies with respect to these background contributions is significant for practial detection of target materials.

This paper develops an independent component analysis (ICA) with subspace projection for hyperspectral anomaly detection. The proposed model represents the sphered data set X as an ICA model specified by a two-component orthogonal sum decomposition, X IC N = +j with j independent components, j IC , generated by ICA and a noise component N. To better extract anomalies from the j IC component space, the concept of sparsity cardinality (SC) is integrated into ICA to derive a ICASC anomaly detector (ICASC-AD). For determine appropriate values of j, the virtual dimensionality (VD) and a minimax-singular value decomposition (MX-SVD) are used for this purpose. The experimental results demonstrate that ICASC are very competitive against the LRaSR-based models in hyperspectral anomaly detection.

A lot of applications as well as in the laboratory range and industrial range need a short-wave infrared imaging system. Especially multispectral imaging in that wavelength region are often enable new applications for quality assurance and monitoring of industrial processes. Due to the cost of the SWIR image sensors a multispectral imaging system should be flexible and adjustable to realize a maximum of applications. Normally push broom devices will fulfill these requirements. Disadvantages are that the sample must be moved, the spectral crosstalk and the blurring between the different channels can disturb the processing, and the correctness of spatial resolution along the scanning direction lead to uncertainties for dimensional measurements. For that reasons a twelve-channel filter-based imaging system was designed. A motor driven filter-wheel with a high precision drive will ensure that the filters will be placed in front of a SWIR Image sensor very precisely. To compensate the spectral aberrations along the optical axis a second drive positioning the image sensor into the focus plane. This enable sharp images in all spectral channels as well as high SNR (depends also on the capability of the SWIR Sensor). Furthermore, this implementation allows also to focus without changing the adjustment of the objective lens. A special designed dome light using halogen bulbs delivers very homogeneous radiation on the sample table. For the image readout, the control of the drives as well as the image presentation a software with a graphical user interface was developed. To export the image stacks in 3rd party software the imaging software saves the multispectral images in the envi-format.

Remote sensing of the Earth allows to receive medium information, a high spatial resolution from space vehicles, and to conduct hyperspectral measurements. This study presents a remote sensing application using time-series Landsat satellite images to monitor the solid waste disposal site (WDS). In this work, neural network image processing methods are used as part of the proposed algorithm for computer modeling of the fractal-percolation process of the underlying surface filtrate in remote sensing imaging. Investigations related to the analysis of the filtrate and decryption of space images are carried out using the apparatus of discrete orthogonal transformations and two convolutional neural networks. The first neural network detects a waste disposal facility, the second works to localize the area of the leachate identified based on the first network. The results obtained can serve as a basis for developing a methodology for assessing the effectiveness of measures to neutralize the underlying surface of waste disposal sites from leachate and its seepage into the soil using remote sensing technologies. This technique can become the object of further research on developing a medical-prophylactic expert system at the territorial level for the detection and neutralization of unauthorized waste disposal sites based on medium and high-resolution space images. As a result, the proposed method demonstrates good accuracy in detection the solid waste disposal site on real satellite images.

Stresses in silicon devices that are introduced during semiconductor processing are a serious concerns in semiconductor fabs. The main sources for stress are doping, surface processing methods and the final chip separation process. Raman spectroscopy is a non-destructive metrology tool that can measure stresses in silicon. Here we present results of Raman analysis that were performed along the different stages of the semiconductor processing chain. In this paper we focused on surface processing and the chip separation process. Our investigations show that micro-Raman spectroscopy is a powerful tool for measuring stress levels and distributions quantitatively on entire productive wafers as well as on the final chip.

Due to the high-dimensional nature of hyperspectral images, dimensionality reduction techniques can improve image analysis by reducing memory requirements, making image processing techniques more efficient, and sometimes increasing the accuracy of the results. A wide variety of dimensionality reduction (DR) algorithms have been proposed, ranging from linear transformations like Principal Component Analysis (PCA) to more computationally intensive manifold learning algorithms such as Locally Linear Embedding (LLE). The best dimensionality reduction technique and the ideal dimensionality depend on the dataset and the image processing task under consideration. This paper compares various dimensionality reduction methods by evaluating the performance of image analysis tasks on available hyperspectral datasets with ground truth reference.