The capability to generate realistic hyperspectral imagery plays a prominent role in applications to sensor and mission development as well as in the training of machine learning models. Yet, it is a challenging task due to the high dimensionality and complex spectral–spatial structure of the data. This paper proposes a novel unsupervised deep-learning framework for generating realistic hyperspectral imagery based on blind hyperspectral unmixing and denoising diffusion probabilistic models. First, the approach extracts both endmembers and abundance maps from hyperspectral data through a dictionary of hyperspectral unmixing algorithms. The extracted abundances are then used as inputs for a guided diffusion model, which serves as the generative framework with the goal of producing realistic synthetic abundance maps. Finally, the generation of synthetic hyperspectral images is accomplished by integrating the generated abundance maps with the extracted endmember set and by suitably conditioning the probabilistic formulation of the guided diffusion model as a function of the unmixing algorithms in the aforementioned dictionary. By combining a collection of blind linear unmixing techniques with the generative capabilities of diffusion models, the proposed methodology aims to address key challenges in simulating hyperspectral sensor outputs. The methodology was validated experimentally using real satellite hyperspectral imagery from the PRISMA mission of the Italian Space Agency. The results confirm the effectiveness of the approach in generating realistic synthetic hyperspectral images associated with various land-covers. The code is available at: https://github.com/martinapastorino/HSI_DDPM.

This paper provides an overview of the main activities and results of HYPERHEALTH project (funded by the Italian Space Agency). Specific focus of this paper is hyperspectral PRISMA data exploitation, mostly as regards PRISMA-based atmospheric constituent estimation and allergenic vegetation monitoring.

PRISMA is a hyperspectral pushbroom sensor, launched by the Italian Space Agency in 2019. PRISMA collects the reflected Earth signal from VNIR to the SWIR with 230 spectral bands with a variable FWHM according to the prism dispersion element. This work intends to develop a procedure suitable to monitor the consistency of photon and thermal noise components across a times series of L1 radiance images collected on different Mediterranean scenarios (i.e. rural and coastal). To improve the retrieval of the useful signal and the random noise on PRISMA images the spatial variability of the scenes has been considered in the new version of the HYperspectral Noise Parameters Estimation (HYNPE) algorithm. The procedure, tested on two PRISMA time series, has assessed quite stable and coherent values for the retrieved noise coefficients, not significantly affected by seasonal radiance variations and scene characteristics

HYPERHEALTH project is co-funded by Italian Space Agency (ASI) in the framework of the "PRISMA Scienza" program. The program supports R&D projects proposed by experts in hyperspectral remote sensing sector from national public research institutions to industries, also in the framework of international partnerships. The aim is designing, developing and testing innovative methods, techniques and algorithms for exploitation of hyperspectral data, with reliable perspectives as to engineering and pre-operational development, thus contributing to the improvement of socio-economic benefits of the end-user community. This paper outlines HYPERHEALTH main goals and activities.

The unique spectral content provided by PRISMA's hyperspectral sensor gives the possibility to study the Earth's surface and environment from space in unprecedented detail. In this respect, our work presents the results of an Italian Space Agency-funded project aiming to develop eight prototypes for providing Value Added products based on such data. Prototypes focus on vegetation, urban areas, water quality, material detection, and natural hazards, combining multiple state-of-the-art techniques based on Machine Learning, physical models, and index-based algorithms. This is particularly relevant given the increasing demand for accurate information to address sustainable policies and support decision-making processes. Through a series of case studies, we highlight the versatility and utility of PRISMA's hyperspectral data for various scientific and operational applications.

The rapid growth of hyperspectral satellite missions and the subsequent availability of hyperspectral images have encouraged the remote sensing community to investigate their potential in estimating the concentration of gases, as CH4and CO2, which are related to the greenhouse effect. Though satellite hyperspectral sensors are not specifically designed for this purpose, they are expected to complement more specific satellite missions, such as NASA’s OCO-2 and OCO-3, both in terms of enriched temporal sampling and improved spatial resolution. In this work, we present a new method to estimate the column-averaged dry-air mole fraction of CO2from hyperspectral data on a per-pixel basis. The method, which is here tailored to PRISMA images, leverages the spectral radiance samples collected in the SWIR spectral region around the CO2absorption band at 2000 nm. By assuming a linear model to describe the dependence of the observed radiance on the CO2concentration, the estimation problem is reduced tomatched filteringand can be effectively implemented in compliance with the low computational burden required to perform a pixel by pixel analysis. The performance of the presented method is investigated by means of a rigorous, physically based simulator that accurately reproduces the at-sensor radiance allowing one to check the validity of the assumptions and to assess the algorithm accuracy. The results show that the presented algorithm outperforms a benchmark CIBR based approach, which has been proposed in the literature to get fast per-pixel estimates of CO2concentration.

In this work, the Learning-Based Approach to Atmospheric Compensation (LBAC) of hyperspectral data proposed by Acitoet al.is extended. LBAC makes use of machine learning methods to directly estimate the spectral reflectance from the at-sensor radiance accounting for the variability induced by one or more unknown atmospheric parameters and by-passing their estimation. LBAC training is obtained by exploiting a spectral reflectance library and accounting for the effects of both the atmosphere and the noise. However, depending on the spectral library adopted, some specific spectra may be reconstructed with lower accuracy. To overcome this drawback, two solutions are proposed referring to two application scenarios. The former deals with small and rare anomalous pixels with unknown reflectance and could be of interest in many applications such as man-made targets detection. It leverages the strengths of LBAC and those of the empirical line method (ELM). The second scenario refers to the case of materials witha prioriknown spectral reflectance and is defined for applications such as mining exploration and contaminant detection. It directly acts on the training phase of LBAC by introducing the spectra of interest in the generation of the training set. An extensive analysis is carried out on simulated data to test the effectiveness of the proposed solutions, to discuss their strengths and weakness, and to compare them with a classical physics-based approach. Results on a real hyperspectral image acquired by an airborne sensor provide a demonstration of the effectiveness of the proposed strategies in a real application environment.

In satellite hyperspectral sensors, a standard procedure based on homogeneous reference sources is used to regularly update the map of defective pixels (DPs). Unfortunately, this procedure often fails to detect some DPs both because they may arise in the interval between two standard calibration steps and because they may be characterized by subtle or unexpected signal values. The resulting hyperspectral image is affected by residual nonuniformity noise which correlates in the along track direction. This noise source reduces the quality of hyperspectral products, such as classification, unmixing and material detection. In this paper, we present a new procedure to find the location of the residual DPs in the detection matrix and we also propose an effective method to estimate the missing radiance values inferring them from the image pixels by leveraging both spatial and spectral correlation. The procedure, here tailored to PRISMA hyperspectral images, is quite general and can be easily adapted to process images recorded by any satellite pushbroom hyperspectral sensor. The improved image quality yielded by the proposed procedure is first demonstratedqualitatively, by comparing the GRX maps on the original image with those obtained after detection of the DPs and correction of their radiance values. The image quality improvement is thenquantifiedon a set of nine PRISMA images, recorded in an interval of about two years, using twoad-hocdefined indexes. The analysis is carried out separately for the VNIR and SWIR spectrometers of the PRISMA mission.

In this work we extend the recently proposed Learning-Based approach to Atmospheric Compensation (LBAC) with respect to two different aspects: 1) the platform for data acquisition and 2) the spectral range covered by the sensor. Particularly, we propose the extension of LBAC to spaceborne hyperspectral sensor operating in the Visible Near InfraRed (VNIR) and Short-Wave InfraRed (SWIR) portion of the electromagnetic spectrum. We specifically refer to the sensor of the PRISMA (PRecursore IperSpettrale della Missione Applicativa) mission, the recent Earth Observation mission of the Italian Space Agency that offers a great opportunity to improve the knowledge on the scientific and commercial applications of spaceborne hyperspectral data. Results obtained on PRISMA hyperspectral images are presented and discussed.

In this work, we deal with the problem of atmospheric compensation (AC) of hyperspectral data collected in the visible and near-infrared (VNIR) spectral range. We propose the “learning-based” approach which uses artificial intelligence algorithms to directly estimate the spectral reflectance from the observed at-sensor radiance image. It uses a parametric regressor whose parameters are learned by means of a strategy based on synthetic data. Such data are generated taking into account 1) the radiative transfer in the atmosphere; 2) the variability of the surface spectral reflectance; and 3) the effects of signal-dependent random noise and spectral miscalibration errors. According to this general framework, we propose a specific multilinear regressor that starting from the knowledge of the atmospheric visibility compensates the water absorption and provides the spectral reflectance of each pixel of the analyzed image. Furthermore, a specific image-based procedure is presented for visibility estimation. The experiment over simulated data is presented and discussed. The test on simulated data aims at showing the effectiveness of the proposed strategy in a completely controlled environment. Experiments are also carried out on three real hyperspectral images acquired by two hyperspectral sensors. The obtained results confirm the effectiveness of the proposed approach by comparing the retrieved reflectance spectra with in-situ measurements or with those obtained by using a well-known commercial AC software.

This letter presents a new method to detect materials with known spectral emissivity in data acquired by longwave infrared hyperspectral sensors. The proposed approach differs from existing methods because it takes into account the uncertainty of the downwelling radiance. Such uncertainty is addressed assuming that the downwelling radiance spans a low-rank subspace whose basis matrix is learned, regardless of the analyzed image, from MODTRAN simulated spectra. The analysis, carried out over data simulated by considering different atmospheric conditions, surface temperatures, and emissivity spectra, shows the effectiveness of the proposed algorithm.

Estimation of the total column water vapor (CWV) content of the atmosphere plays an important role in the atmospheric compensation (AC) of remotely sensed hyperspectral images collected in the visible and near infrared (VNIR) spectral range. Most of the proposed CWV retrieval methods provide accurate estimates as long as other significant atmospheric parameters are known. Those parameters are not generally available and must in turn be estimated. In this article, a new approach based on deep learning is proposed that allows the estimation of CWV without the knowledge of the atmospheric visibility, the solar zenith angle, and the atmospheric point spread function (PSF). The proposed approach includes a training strategy based on synthetic data that are generated according to an accurate radiative-transfer model, and by exploiting reflectance spectral libraries and the MODTRAN radiative-transfer code. Experiments on simulated data are carried out to analyze the performance of the proposed deep neural network with reference to both aerial and satellite applications. Furthermore, an example of the results provided by the network in a real application is shown. For this purpose, the network is applied to data acquired by an airborne hyperspectral sensor operating in the VNIR spectral range.

In this paper, we investigate the temperature and emissivity separation (TES) problem from hyperspectral data acquired in the long-wave infrared region (LWIR) of the electromagnetic spectrum. We derive a general class of TES algorithms [subspace-based TES (SBTES)] relying on the assumption that the emissivity spectra of natural and man-made materials can be well represented in a given subspace of the original data space. Specifically, by exploiting the subspace representation and the Gaussian model for the noise affecting LWIR hyperspectral data, we approach TES under a statistical perspective by obtaining the maximum likelihood estimates of both the temperature and the spectral emissivity. The proposed approach originates several algorithms whose specific form depends on the particular basis matrix adopted to address the emissivity subspace. We study the performance of the presented class of algorithms and derive theoretical bounds on the accuracy of the temperature and emissivity estimators. Furthermore, by specifying two basis matrices for the emissivity subspace, we propose two different algorithms within the SBTES class. Finally, we present the results of an extensive experimental analysis carried out over simulated data to assess and compare the performance of the two presented algorithms.

This paper deals with atmospheric correction in hyperspectral data acquired in the long-wave infrared (LWIR) spectral range. Atmospheric compensation (AC) is approached from a new perspective in that it is reformulated as an estimation problem on a low-rank subspace. Specifically, a subspace-based model is exploited to represent both the atmospheric transmittance and the upwelling radiance. Taking into account the inherent correlation between these two quantities, we adopt a subspace model that constrains them to vary in a physically consistent way. Exploiting such a model, AC is formulated as a nonlinear least-squares estimation problem. An automatic procedure is proposed to estimate the unknown parameters, which is based only on the image analysis and does not need any atmospheric measurements. Results of an extensive experimental analysis carried out on simulated data are used to discuss the performance of the algorithm with respect to different atmospheric conditions. Finally, experimental evidence over a real LWIR hyperspectral image is provided.

Atmospheric compensation (AC) is a fundamental and critical step for quantitative exploitation of hyperspectral data. It is the means by which the reflectance of an object/material is estimated from the measured at-sensor radiance. Such reflectance is the inherent signature that is used to identify various materials in a monitored scene. AC is quite complex and is hampered by the large amount of uncontrollable variables that play a role: just think about the spatial variability of some atmospheric constituents such as water vapor and aerosols, or to the rapidly spatially varying effects of the radiation coming from adjacent areas. Though, in principle, some atmospheric parameters and radiometric quantities such as solar irradiance and sky irradiance can be measured during the flight, in practice such measures are rarely available in an operational framework or are taken at a single point of the surface ignoring their spatial variation. Thus, a prompt quantitative exploitation of hyperspectral data for operational purposes, such as material identification and object detection, requires unsupervised and accurate AC procedures that can learn from the image itself the parameters of the inversion model and follow their variability within the scene. In this framework, we present a new unsupervised methodology for AC of airborne hyperspectral images in the visible and near-infrared spectral range. The proposed methodology relies on a radiative transfer model accounting for the adjacency effect and allows the estimation of relevant atmospheric parameters. Specifically, it embeds two new algorithms for the estimation of: 1) aerosol and atmospheric visibility and 2) the water vapor content of the atmosphere accounting for the spatial variability of such a parameter. The two algorithms significantly differ from those adopted by existing state-of-the-art approaches or in commercial packages such as fast line-of-sight atmospheric analysis of spectral hypercubes and airborne atmospheric and topographic correction algorithm. In this paper, we present the detailed description of the new AC methodology, and we analyze the results provided by the algorithm over real data.

The knowledge of atmospheric column water vapor concentration is crucial for compensating water absorption effects in remote sensing data. Several algorithms for the estimation of such a parameter were proposed in the past. One of the most effective algorithms is the atmospheric precorrected differential absorption (APDA) technique. APDA relies on a simplified radiative transfer model (RTM) that does not account for the spatial variability of the adjacency effects. In this paper, we study the impact of the simplified RTM assumption on the performance of the algorithm by exploiting a more realistic and well-established RTM. Starting from such a model, we derive a new water retrieval algorithm called low-rank subspace projection-based water estimator. It exploits the high degree of spectral correlation experienced in the reflectances of most of the existing materials. An extensive experimental analysis is carried out on simulated data in order to assess and compare the performance of the two algorithms. Simulation results allow the critical analysis of the two algorithms by highlighting their strengths and drawbacks.

An experimental study is presented with the goal of investigating application of the forward modeling (FM) approach to target detection to hyperspectral images acquired over an urban scenario, where the effects of variability and uncertainty of atmospheric and illumination conditions are enhanced. The study exploits the availability of a new hyperspectral data set acquired in May 2013 over the city of Viareggio, Italy. A physics-based model is assumed for radiation transfer in the atmosphere and variability around nominal acquisition conditions is accounted for by means of multiple executions of a radiative transfer code and a subspace-based target detection framework. Experimental results featuring several ground-truthed targets in an urban scenario confirm the effectiveness of the FM target detection approach.

This paper presents a new procedure to make the at-sensor radiance comparable with the object reflectance spectrum in order to perform object detection and identification. A new transformation is proposed to account for the non-perfect knowledge of illumination, viewing and atmospheric conditions. The transformation, which is applied to both the image data and the object spectrum, takes into account the uncertainty of the acquisition conditions by resorting to a subspace based approach. The subspace is obtained by using a parametric physics-based radiative transfer model and by resorting to the MODerate resolution TRANmission (MODTRAN) radiative transfer code. The procedure is entailed in a target detection and identification scheme and is applied to real hyperspectral data acquired during a recently performed measurement campaign.

This paper presents a multi-temporal approach to compensate the atmospheric effects in spaceborne hyperspectral images. It focuses on applications where a hyperspectral sensor periodically acquires images of the same scene and such images have to be classified. The method assumes that a reference reflectance image of the region of interest, equipped with an accurate classification map, is available. Such an image and the corresponding classification map represent the database that will be a reference for the classification of temporally subsequent images over the same scene. The multi-temporal approach allows the classification to be performed on at sensor radiance data and it does not require the knowledge of the atmospheric conditions at the acquisition time.

The SAP4PRISMA project research activities aimed at supporting the Italian hyperspectral PRISMA mission by developing preliminary processing chains suitable for PRISMA to obtain high level hyperspectral data products for agriculture, land degradation, natural and human hazards.

The main objectives of the PRISMA (Hyperspectral Precursor of the Application Mission) mission are: the implementation of an Earth Observation pre-operative payload, the in-orbit demonstration and qualification of an Italian state-of-the-art hyperspectral/panchromatic technology and the validation of end-to-end data processing system able to support the development of new applications based on high spectral resolution images. The aim of the paper is to provide an overview of the PRISMA mission by describing the current status of the program and giving a brief outline of the work done till now in the framework of the SAP4PRISMA project scientific studies in supporting the exploitation of the future PRISMA hyperspectral images for environmental applications.

Orthogonal subspace projection (OSP) is a powerful tool for dimensionality reduction (DR) in hyperspectral images (HSIs). In the OSP approach, the basis of the signal subspace must be estimated from the data themselves. Such estimation task is referred to as signal subspace identification (SSI). Most of the SSI methods in the literature are based on the analysis of the data second-order statistics (SOS) and have two main drawbacks: 1) They do not take into account the rare signal components (or rare vectors), 2) they assume that noise is spatially stationary. Rare vectors are those signal components that are present in pixels scarcely represented in the image and linearly independent on the signal components characterizing the rest of the image pixels. SOS-based SSI algorithms estimate the signal subspace addressing mostly the background and ignoring the presence of the rare pixels. This may be detrimental for the performance of detection algorithms when DR is adopted as a pre-processing step in small target detection applications. In this paper, a new technique for SSI in HSIs is presented. The algorithm is developed to account for both the abundant and the rare signal components. The method is derived by assuming a signal-dependent model for the noise affecting the data. This makes the SSI algorithm particularly suitable for the processing of images acquired by new generation sensors where, due to the improved sensitivity of the electronic components, noise includes a signal-dependent term. Results on simulated data are discussed, and the comparison with a recently proposed technique based on the analysis of SOS is performed. Furthermore, the results obtained by applying the SSI algorithm to a real HSI affected by signal-dependent noise are presented and discussed.

In this paper, we investigate the constant false-alarm rate (CFAR) property of the RX anomaly detector which is widely used for the analysis of hyperspectral data. The RX detector relies on an adaptive scheme where the mean vector and the covariance matrix of the background are locally estimated from the image pixels themselves. First, demeaning is accomplished by removing the estimated local background mean value, and then, the covariance matrix is estimated in a homogeneous neighborhood of each pixel. In principle, if the local mean is perfectly removed and the covariance matrix is estimated from background pixels sharing the same covariance matrix, the RX algorithm has the CFAR property, which is highly desirable in practical applications. The CFAR behavior of the algorithm also requires the spatial stationarity of the random noise affecting the hyperspectral image. In data collected by new-generation sensors, such an assumption is not valid because photon noise contribution, which depends on the spatially varying signal level, is not negligible. This has motivated us to analyze the behavior of the RX algorithm with respect to the CFAR property in data affected by signal-dependent (SD) noise. In this paper, we show both theoretically and experimentally that the SD noise is one of the causes of the non-CFAR behavior of the RX detector that we have experienced in many practical situations. We propose a strategy to enhance the robustness of the anomaly detection scheme with respect to the CFAR property based on an adaptive nonlinear transform aimed at reducing the dependence of the noise on the signal level. Experiments on simulated data and real data collected by a new hyperspectral camera are also presented and discussed.

In this paper we investigate the effects of the signal dependent noise on the CFAR property of the RX anomaly detector. The CFAR behaviour of the algorithm was proved under the assumption of spatial stationarity of the random noise affecting the hyperspectral image. In data collected by new generation sensors such an assumption is not valid because photon noise contribution, which depends on the spatially varying signal levels, is not negligible. In this paper, experiments on real data collected by a new hyperspectral camera are discussed in order to show that the signal dependent noise is one of the causes of the non-CFAR behaviour of the RX detector we have experienced in many practical situations.

The SAP4PRISMA is a four year research project which aims at developing algorithms and products for the future PRISMA mission. The project started on May 2010 and is now entering his full activities as the ”PRISMA like” data set has been defined and the test areas were selected. The paper describes the main project objectives and the activities realized in the first 9 months of the project.

In the framework of the monitoring of structures and infrastructures from environmental disasters, the COSMO-SkyMed constellation has a huge potential, thanks to up to metric spatial resolution, short revisit time, and the day/night all-weather acquisition capability ensured by SAR. This paper focuses on the scientific results of the project “Development and validation of multitemporal image analysis methodologies for multirisk monitoring of critical structures and infrastructures,” funded by the Italian Space Agency. Several change-detection, data-fusion, and feature-extraction techniques, which were developed and experimentally validated in the project for COSMO-SkyMed imagery and for their integration with other data sources (including very high resolution optical data), are described and examples of processing results are discussed.

In this paper, a new algorithm for striping noise reduction in hyperspectral images is proposed. The new algorithm exploits the orthogonal subspace approach to estimate the striping component and to remove it from the image, preserving the useful signal. The algorithm does not introduce artifacts in the data and also takes into account the dependence on the signal intensity of the striping component. The effectiveness of the algorithm in reducing striping noise is experimentally demonstrated on real data acquired both by airborne and satellite hyperspectral sensors.

In this paper, a novel method to characterize random noise sources in hyperspectral (HS) images is proposed. Noise is described using a parametric model that accounts for the dependence of noise variance on the useful signal. Such model takes into account the photon noise contribution and is therefore suitable for noise characterization in the data acquired by new-generation HS sensors where electronic noise is not dominant. A new algorithm is developed for the estimation of noise parameters which consists of two steps. First, the noise and signal realizations are extracted from the original image by resorting to the multiple-linear-regression-based approach. Then, the model parameters are estimated by using a maximum likelihood approach. The new method does not require the intervention of a human operator and the selection of homogeneous regions in the scene. The performance of the new technique is analyzed on simulated HS data. Results on real data are also presented and discussed. Images acquired with a new-generation HS camera are analyzed to give an experimental evidence of the dependence of random noise on the signal level and to show the results of the estimation algorithm. The algorithm is also applied to a well-known Airborne Visible/Infrared Imaging Spectrometer data set in order to show its effectiveness when noise is dominated by the signal-independent term.

This paper deals with subspace-based target detection in hyperspectral images. Specifically, it focuses on a general detection scheme where, first, background is suppressed through orthogonal-subspace projection and then target detection is accomplished. An adequate estimation of the background subspace is essential to a successful outcome. The background subspace has been typically estimated globally. However, global approaches may be ineffective for small-target-detection applications since they tend to overestimate the background interference affecting a given target. This may result in a low target residual energy after background suppression that is detrimental to detection performance. In this paper, we propose a novel and fully automatic algorithm for local background-subspace estimation (LBSE). Local background has typically a lower inherent complexity than that of global background. By estimating the background subspace over a local neighborhood of the test pixel, the resulting background-subspace dimension is expected to be low, thus resulting in a higher target residual energy after suppression which benefits the detection performance. Specifically, the proposed LBSE acts on a per-pixel basis, thus adaptively tailoring the estimated basis to the local complexity of background. Both simulated and real hyperspectral data are employed to investigate the detection-performance improvements offered by LBSE with respect to both global and local methodologies previously presented.

In this paper, we investigate the problem of signal subspace identification (SSI) and dimensionality reduction in hyperspectral images. We consider two recently proposed SSI algorithms: the Maximum Orthogonal Complement Analysis (MOCA) algorithm and the Robust Signal Subspace Estimator (RSSE) algorithm. Such algorithms are robust to the presence of rare signal components and are particularly effective in reducing the number of features in the preprocessing step for small target detection applications. In this paper, MOCA and RSSE are briefly revisited and integrated in a common theoretical framework in order to better highlight and understand their peculiarities. Furthermore, their performances are compared in terms of computational complexity and of their ability to address both the abundant and the rare signal components. A modified version of the MOCA is also introduced, which is computationally more efficient than the original algorithm. Results on simulated data are discussed, and a case study is presented concerning real Airborne Visible/Infrared Imaging Spectrometer data.

This paper deals with the problem of signal subspace estimation for dimensionality reduction (DR) in hyperspectral images in the presence of rare pixels, i.e., pixels that are scarcely represented in the image and containing spectral components that are linearly independent of the background. Most of the classical methods proposed in the literature are based on the analysis of second-order statistics (SOS), which are weakly influenced by the rare signals. Therefore, such techniques estimate the signal subspace addressing mostly the background and ignoring the presence of rare pixels. This may reduce the target/background spectral contrast, thus decreasing the detection performance when DR is adopted as preprocessing task in small-target detection applications. In this paper, a new robust algorithm, namely, robust signal subspace estimation (RSSE), is developed, which preserves both abundant and rare signal components. It combines the analysis of SOS and a recent approach based on the analysis of thel2infinnorm. The novel contribution of this paper is twofold. First, the RSSE algorithm is presented, which includes a new iterative procedure to derive the signal subspace and an original statistical method to estimate the data dimensionality. Second, anad hocsimulation strategy is proposed to assess the performance of signal subspace estimation methods in the presence of rare signal components. The procedure is adopted to compare the RSSE algorithm with a classical technique based on the analysis of SOS. The results obtained by applying the two methods on a real Airborne Visible Infrared Imaging Spectrometer hyperspectral image are also presented and discussed.

This paper deals with the problem of signal subspace estimation and dimensionality reduction (DR) in hyperspectral images. A new algorithm is presented which preserves both the abundant and the rare signal components and is therefore suitable for DR in target detection applications. Results obtained by applying the new procedure and a classical method based on the analysis of the second order statistics are presented and discussed with reference to real AVIRIS data.

In this manuscript we investigate the efficient implementation of anomaly detection strategies in hyperspectral images. We especially focus on methods to reduce the computational complexity for a fast implementation of the detection algorithms. In particular, we consider two strategies based on data fusion methods applied to the outputs of the optical heads of the hyperspectral sensor. Furthermore, we consider, two computationally efficient implementations of anomaly detection where the well known RX algorithm is applied to hyperspectral data after dimensionality reduction. The detection performances of the anomaly detection strategies are compared using real data acquired by the MIVIS sensor. An estimate of the reduction of the computational load achieved with the different techniques is also provided.

Inverse synthetic aperture radar (ISAR) images are frequently used in target classification and recognition applications. Some classifiers often require features that can be more easily obtained by extracting scattering centers from ISAR data rather than by reconstructing ISAR images. An available method for scattering center extraction, namely, the CLEAN technique, was proposed in a recent paper by Yanget al. In this paper, an improvement of this CLEAN technique is proposed that introduces a new method for detecting scattering centers. The proposed technique is based on a Gaussianity test, and its effectiveness is first theoretically proven and then tested on real data. Moreover, a comparison with the technique proposed by Yanget al. is shown.

In this paper a new strategy aimed at reducing the computational complexity in hyperspectral anomaly detection is introduced. It is based on the fusion of the results obtained by applying the RX detector to the data measured by the different optical systems in the adopted hyperspectral sensor. Two feature level fusion criteria are derived and the computational complexity of each of them is evaluated. A comparison among the RX algorithm detection performance and the ones of the proposed anomaly detectors is provided by considering a data set acquired by an airborne hyperspectral sensor.

In this paper a new sub-pixel target detector for hyper-spectral images, based on the stochastic mixing model (SMM), is presented. The SMM models a mixed pixel, under the target present hypothesis, as linear combination of target and background spectra. Unlike the linear mixing model (LMM), target and background are modeled as random vectors in order to characterize their spectral variability. By assuming the target spectrum deterministic and known, a SMM based detection strategy is derived by computing the least square mean error (LSME) estimate of the target fraction in the observed pixel. This approach provides a closed form detector statistic as opposite to other SMM based detectors proposed in the literature. The new algorithm and the adaptive matched subspace detector (AMSD), based on the LMM, are applied to a MIVIS data set and the experimental results are compared by means of a suitable performance index.

This work deals with the problem of background removal in infra-red (IR) image sequences. Background removal is a basic step to detect small targets in surveillance systems based on IR images. The paper refers to a general background removal procedure that consists in estimating and subtracting the background in each frame of the IR sequence. The estimation step is accomplished by means of linear and non linear filters. The focus of this work is on techniques adopting four different filters: 1) the 2D window average filter; 2) the 2D MEDIAN filter; 3) the max/median filter; 4) the max/mean filter. In the paper a methodology to experimentally compare the performance of the different techniques is described and the results obtained over real IR data are discussed.

In this manuscript we investigate on the statistical modeling of hyperspectral data. Accurately modeling real data is of paramount importance in the design of optimal classification or detection strategies and in evaluating their performances. In the work three nonGaussian models are considered and their capability in characterizing the statistical behavior of real data is discussed with reference to a data set acquired by the multispectral infrared and visible imaging spectrometer (MIVIS) sensor.

This paper deals with the sub-pixel target detection problem in hyper-spectral images. The problem is approached by modeling the mixed spectrum with both the linear mixing model (LMM) and the stochastic mixing model (SMM). A detection strategy is derived by assuming the SMM. In the proposed algorithm, detection is accomplished by testing the values of the maximum a-priori probability (MAP) estimate of the target's abundance that represent the fraction of the spectrum in the observed pixel due to the target. The algorithm has been applied to experimental images and the results have been compared with the ones obtained by the adaptive matched subspace detector (AMSD) based on the LMM

A new algorithm for hyperspectral image segmentation based on the statistical approach is presented. The algorithm is completely unsupervised and relies only on the spectral information. The hyperspectral image is statistically characterized by means of the Gaussian Mixture Model (GMM). Preliminary results obtained on experimental data are presented and discussed.

In this paper we present an automatic procedure for the detection of long-range airborne targets in maritime naval surveillance systems. The algorithm performance is investigated via simulation and compared to the one of the classical moving window detection algorithm.

An efficient algorithm for endmember selection is illustrated. Endmembers are estimated by an unsupervised segmentation procedure based on spectral analysis. Preliminary results obtained on experimental data are presented and discussed.