Yannick Berthoumieu - Academia.edu (original) (raw)
Papers by Yannick Berthoumieu
arXiv (Cornell University), Sep 28, 2019
Most superpixel algorithms compute a trade-off between spatial and color features at the pixel le... more Most superpixel algorithms compute a trade-off between spatial and color features at the pixel level. Hence, they may need fine parameter tuning to balance the two measures, and highly fail to group pixels with similar local texture properties. In this paper, we address these issues with a new Texture-Aware SuperPixel (TASP) method. To accurately segment textured and smooth areas, TASP automatically adjusts its spatial constraint according to the local feature variance. Then, to ensure texture homogeneity within superpixels, a new pixel to superpixel patch-based distance is proposed. TASP outperforms the segmentation accuracy of the state-of-the-art methods on texture and also natural color image datasets.
HAL (Le Centre pour la Communication Scientifique Directe), Jul 23, 2020
The pansharpening problem amounts to fusing a high-resolution panchromatic image with a lowresolu... more The pansharpening problem amounts to fusing a high-resolution panchromatic image with a lowresolution multispectral image so as to obtain a high-resolution multispectral image. So the preservation of the spatial resolution of the panchromatic image and the spectral resolution of the multispectral image are of key importance for the pansharpening problem. To cope with it, we propose a new method based on a bi-discriminator in a Generative Adversarial Network (GAN) framework. The first discriminator is optimized to preserve textures of images by taking as input the luminance and the near infrared band of images, and the second discriminator preserves the color by comparing the chroma components Cb and Cr. Thus, this method allows to train two discriminators, each one with a different and complementary task. Moreover, to enhance these aspects, the proposed method based on bi-discriminator, and called MDSSC-GAN SAM, considers a spatial and a spectral constraints in the loss function of the generator. We show the advantages of this new method on experiments carried out on Pléiades and World View 3 satellite images.
This paper presents a new approach for unsupervised band selection in the context of hyperspectra... more This paper presents a new approach for unsupervised band selection in the context of hyperspectral imaging. The hyperspectral band selection (HBS) task is considered as a clustering problem: bands are clustered in the image space; one representative image is then kept for each cluster, to be part of the set of selected bands. The proposed clustering method falls into the family of information-maximization clustering, where mutual information between data features and cluster assignments is maximized. Inspired by a clustering method of this family, we adapt it to the HBS problem and extend it to the case of multiple image features. A pixel selection step is also integrated to reduce the spatial support of the feature vectors, thus mitigating the curse of dimensionality. Experiments with different standard data sets show that the bands selected with our algorithm lead to higher classification performance, in comparison with other state-of-the-art HBS methods.
2022 IEEE International Conference on Image Processing (ICIP), Oct 16, 2022
Compared to standard deep convolutional neural networks (CNN) which include a global average pool... more Compared to standard deep convolutional neural networks (CNN) which include a global average pooling operator, second-order neural networks have a global covariance pooling operator which allows to capture richer statistics of CNN features. They have been shown to improve representation and generalization abilities. However, this covariance pooling is performed only on the deepest CNN feature maps. To benefit from different levels of abstraction, we propose to extend these models by using a multi-layer approach. In addition, to obtain better predictive performance, an end-to-end ensemble learning architecture is proposed. Experiments are conducted on four datasets and have confirmed the potential of the proposed model for various image processing applications such as remote sensing scene classification, indoor scene recognition and texture classification.
IEEE Transactions on Geoscience and Remote Sensing, Sep 1, 2019
This paper presents a new method to perform unsupervised band selection (UBS) with hyperspectral ... more This paper presents a new method to perform unsupervised band selection (UBS) with hyperspectral data. The method provides a probabilistic clustering approach. The band images are clustered in the image space by computing their posterior class probability. Then, for each cluster, the band exhibiting the highest probability of belonging to it is selected as cluster exemplar. More particularly, the proposed method falls into information-maximization clustering methods, where the posterior class probability is modeled and the parameters of the models are derived by maximizing the information between the data and the unknown cluster labels. In this context, we propose a new image representation for hyperspectral images, based on the first and second order statistics of multiple image features. We refer to this representation as multiple-feature local statistical descriptors (MLSD). The descriptors are computed w.r.t. regular grids, and a special pixel selection procedure reduces the number of samples within each block of the grid. A kernel-based model that embeds the MLSD is then proposed for the posterior class probability. The model is finally optimized according to an information-maximization criterion. We conduct several experiments to determine the best parameters for the proposed approach and compare the latter with other state-of-the-art UBS methods. Quantitative evaluations show that, by employing our band selection method, higher performance in terms of classification accuracy and endmember extraction can be achieved in comparison with the state of the art.
Journal of the Optical Society of America, Oct 23, 2019
2022 30th European Signal Processing Conference (EUSIPCO)
Cet article presente une nouvelle architecture hybride basee sur l'encodage par vecteurs de F... more Cet article presente une nouvelle architecture hybride basee sur l'encodage par vecteurs de Fisher (VF) des sorties des couches convolutives d'un reseau de neurones. L'originalite de ce travail repose sur l'exploitation des statistiques d'ordre deux via le calcul des matrices de covariance locales. Considerant les proprietes intrinseques a la geometrie Riemannienne propre a l'espace des matrices de covariance, nous proposons d'utiliser la metrique log-euclidienne afin d'etendre le concept des VF pour l'encodage de matrices de covariance : les vecteurs de Fisher log-euclidiens (LE VF). L'architecture proposee est ensuite evaluee sur differentes bases de donnees de teledetection : la base UC Merced Land Use Land Cover, la base AID, ainsi que sur deux jeux de donnees Pleiades sur des forets de pins maritimes et de parcs ostreicoles.
In this paper, we introduce a new generative adversarial network (GAN) with dual image-color disc... more In this paper, we introduce a new generative adversarial network (GAN) with dual image-color discriminators, to predict Artificial-SAR colorized images from SAR ones (Sentinel-1). Based on the conventional architecture of GANs, we employ an additional color discriminator that evaluates the differences in brightness, contrast, and major colors between images, while the image discriminator compares texture and content. To achieve the required level of colorization in the generation process, we employ non-adversarial color loss dedicated for color comparison, unlike conventional approaches that use only L1 loss. Moreover, to overcome the vanishing gradient problem in deep architecture, and ensure the flow of low-level information inside network layers, we add residual connections to our generator that follows the general shape of U-Net. The performance of the proposed model was evaluated quantitatively as well as qualitatively with the SEN1-2 dataset. Results show that the proposed mod...
2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Mar 1, 2017
Many signal and image processing applications, including SAR polarimetry and texture analysis, re... more Many signal and image processing applications, including SAR polarimetry and texture analysis, require the classification of complex covariance matrices. The present paper introduces a geometric learning approach on the space of complex covariance matrices based on a new distribution called Riemannian Gaussian distribution. The proposed distribution has two parameters, the centre of massȲ and the dispersion parameter σ. After having derived its maximum likelihood estimator and its extension to mixture models, we propose an application to texture recognition on the VisTex database.
IEEE Transactions on Image Processing, 2014
This index covers all technical items-papers, correspondence, reviews, etc.-that appeared in this... more This index covers all technical items-papers, correspondence, reviews, etc.-that appeared in this periodical during 2014, and items from previous years that were commented upon or corrected in 2014. Departments and other items may also be covered if they have been judged to have archival value. The Author Index contains the primary entry for each item, listed under the first author's name. The primary entry includes the coauthors' names, the title of the paper or other item, and its location, specified by the publication abbreviation, year, month, and inclusive pagination. The Subject Index contains entries describing the item under all appropriate subject headings, plus the first author's name, the publication abbreviation, month, and year, and inclusive pages. Note that the item title is found only under the primary entry in the Author Index.
Journal of Applied Geophysics, 2018
For a geoscientist, the Relative Geologic Time (RGT) is an important tool to perform chronostrati... more For a geoscientist, the Relative Geologic Time (RGT) is an important tool to perform chronostratigraphic analysis. However, automatically estimate an RGT image from a seismic image can be a challenging task where we have to respect seismic features, the deposit orders and to deal efficiently with unconformities such as erosions, progradating systems, etc. To this end, approaches have been proposed formulating the estimation problem in a regularized convex optimization problem. However none of these fully address efficiently all issues. In this paper, we propose a new regularization term based on an asymmetric and adaptive weight function. The asymmetric behavior focuses on the ill-posed problem while the adaptive process is used to deal with unconformities by modulating the strength of the regularization if necessary. Moreover, to increase the robustness of the approach, we propose variants of the method in terms of ℓ 1-norm instead of ℓ 2-norm that corresponds to potentially too smooth solutions. For evaluating the relevance of our proposals, experimentations have been conducted on both synthetic and real seismic images.
Ce travail est parrainé par le Gdr Automatique « la dérivation non entière en isolation vibratoir... more Ce travail est parrainé par le Gdr Automatique « la dérivation non entière en isolation vibratoire » Résumé-Dans un contexte de synthèse de profil routier, cet article présente l'extension d'une méthode de synthèse monodimensionnelle à la génération d'un terrain fractal auto-similaire 2-D. L'objectif est de fournir un terrain dont la dimension fractale est isotrope en terme d'évolution spatiale. La méthode de synthèse s'appuie sur une décomposition de Cholesky. Nous montrons en outre le lien qu'il existe entre les matrices issues de la décomposition selon l'élongation longitudinale et transversale.
Physics in Medicine and Biology, 2006
Non-invasive methods for quantifying [(18)F]FDG uptake in tumours often require normalization to ... more Non-invasive methods for quantifying [(18)F]FDG uptake in tumours often require normalization to either body weight or body surface area (BSA), as a surrogate for [(18)F]FDG distribution volume (DV). Whereas three dimensions are involved in DV and weight (assuming that weight is proportional to volume), only two dimensions are obviously involved in BSA. However, a fractal geometry interpretation, related to an allometric scaling, suggests that the so-called 'body surface area' may stand for DV.
This paper presents a novel framework for visual content classification using jointly local mean ... more This paper presents a novel framework for visual content classification using jointly local mean vectors and covariance matrices of pixel level input features. We consider local mean and covariance as realizations of a bivariate Riemannian Gaussian density lying on a product of submanifolds. We first introduce the generalized Mahalanobis distance and then we propose a formal definition of our product-spaces Gaussian distribution on R m × SPD(m). This definition enables us to provide a mixture model from a mixture of a finite number of Riemannian Gaussian distributions to obtain a tractable descriptor. Mixture parameters are estimated from training data by exploiting an iterative Expectation-Maximization (EM) algorithm. Experiments in a texture classification task are conducted to evaluate this extended modeling on several color texture databases, namely popular Vistex, 167-Vistex and CUReT. These experiments show that our new mixture model competes with state-of-the-art on the experimented datasets.
arXiv (Cornell University), Jun 7, 2022
-Les méthodes de super résolution d'image visentà recréer une image haute résolutionà partir d'un... more -Les méthodes de super résolution d'image visentà recréer une image haute résolutionà partir d'une basse résolution. La famille d'approches basée sur les patchs a fait l'objet d'une attention et d'un développement considérable. La technique de minimum de l'erreur quadratique moyenne est une méthode de restauration d'images qui utilise un modèle gaussien de probabilités sur les patchs d'images. Cet article propose un algorithme d'apprentissage d'un modèle conjoint de mélange gaussien généralisé (GGMM)à partir des paires de patchsà basse résolution et des patchs correspondantsà haute résolution d'une image de référence.À partir de ce modèle GGMM, l'image haute résolution en utilisant la méthode MMSE. Nosévaluations numériques indiquent que la méthode MMSE-GGMM se comporte très bien par rapportà l'état de l'art.
Pattern Recognition, Oct 1, 2023
Lecture Notes in Computer Science, 2022
IEEE Transactions on Geoscience and Remote Sensing, Aug 1, 2016
In order to obtain accurate classification results of hyperspectral images, both the spectral and... more In order to obtain accurate classification results of hyperspectral images, both the spectral and spatial information should be fully exploited in the classification process. In this paper, we propose a novel method using independent component analysis (ICA) and edge-preserving filtering (EPF) via an ensemble strategy for the classification of hyperspectral data. First, several subsets are randomly selected from the original feature space. Second, ICA is used to extract spectrally independent components followed by an effective EPF method, to produce spatial features. Two strategies (i.e., parallel and concatenated) are presented to include the spatial features in the analysis. The spectralspatial features are then classified with a random forest (RF) or rotation forest (RoF) classifier. Experimental results on two real hyperspectral datasets demonstrate the effectiveness of the proposed methods. A sensitivity analysis of the new classifiers is also performed. Index Terms Classification, hyperspectral data, independent component analysis (ICA), edge preserving filter (EPF). I. INTRODUCTION During the past two decades, the development of hyperspectral sensors have resulted in great improvement for the image acquisition capabilities. Hyperspectral sensors are now able to provide images with both high spectral and spatial resolutions. Hence, hyperspectral data offers a unique opportunity to monitor the Earth surface. Thematic applications include environmental mapping and crop monitoring [1], [2]. Supervised classification is one of the most important problems in the remote sensing community. Given a set of training samples (i.e., pixel vectors for hyperspectral image), the aim of classification is to assign a unique Manuscript received ; revised. This study has been carried out with financial support from the French State, managed by the French National Research Agency (ANR) in the frame of the "Investments for the future" Programme IdEx Bordeaux-CPU (ANR-10-IDEX-03-02).
arXiv (Cornell University), Sep 28, 2019
Most superpixel algorithms compute a trade-off between spatial and color features at the pixel le... more Most superpixel algorithms compute a trade-off between spatial and color features at the pixel level. Hence, they may need fine parameter tuning to balance the two measures, and highly fail to group pixels with similar local texture properties. In this paper, we address these issues with a new Texture-Aware SuperPixel (TASP) method. To accurately segment textured and smooth areas, TASP automatically adjusts its spatial constraint according to the local feature variance. Then, to ensure texture homogeneity within superpixels, a new pixel to superpixel patch-based distance is proposed. TASP outperforms the segmentation accuracy of the state-of-the-art methods on texture and also natural color image datasets.
HAL (Le Centre pour la Communication Scientifique Directe), Jul 23, 2020
The pansharpening problem amounts to fusing a high-resolution panchromatic image with a lowresolu... more The pansharpening problem amounts to fusing a high-resolution panchromatic image with a lowresolution multispectral image so as to obtain a high-resolution multispectral image. So the preservation of the spatial resolution of the panchromatic image and the spectral resolution of the multispectral image are of key importance for the pansharpening problem. To cope with it, we propose a new method based on a bi-discriminator in a Generative Adversarial Network (GAN) framework. The first discriminator is optimized to preserve textures of images by taking as input the luminance and the near infrared band of images, and the second discriminator preserves the color by comparing the chroma components Cb and Cr. Thus, this method allows to train two discriminators, each one with a different and complementary task. Moreover, to enhance these aspects, the proposed method based on bi-discriminator, and called MDSSC-GAN SAM, considers a spatial and a spectral constraints in the loss function of the generator. We show the advantages of this new method on experiments carried out on Pléiades and World View 3 satellite images.
This paper presents a new approach for unsupervised band selection in the context of hyperspectra... more This paper presents a new approach for unsupervised band selection in the context of hyperspectral imaging. The hyperspectral band selection (HBS) task is considered as a clustering problem: bands are clustered in the image space; one representative image is then kept for each cluster, to be part of the set of selected bands. The proposed clustering method falls into the family of information-maximization clustering, where mutual information between data features and cluster assignments is maximized. Inspired by a clustering method of this family, we adapt it to the HBS problem and extend it to the case of multiple image features. A pixel selection step is also integrated to reduce the spatial support of the feature vectors, thus mitigating the curse of dimensionality. Experiments with different standard data sets show that the bands selected with our algorithm lead to higher classification performance, in comparison with other state-of-the-art HBS methods.
2022 IEEE International Conference on Image Processing (ICIP), Oct 16, 2022
Compared to standard deep convolutional neural networks (CNN) which include a global average pool... more Compared to standard deep convolutional neural networks (CNN) which include a global average pooling operator, second-order neural networks have a global covariance pooling operator which allows to capture richer statistics of CNN features. They have been shown to improve representation and generalization abilities. However, this covariance pooling is performed only on the deepest CNN feature maps. To benefit from different levels of abstraction, we propose to extend these models by using a multi-layer approach. In addition, to obtain better predictive performance, an end-to-end ensemble learning architecture is proposed. Experiments are conducted on four datasets and have confirmed the potential of the proposed model for various image processing applications such as remote sensing scene classification, indoor scene recognition and texture classification.
IEEE Transactions on Geoscience and Remote Sensing, Sep 1, 2019
This paper presents a new method to perform unsupervised band selection (UBS) with hyperspectral ... more This paper presents a new method to perform unsupervised band selection (UBS) with hyperspectral data. The method provides a probabilistic clustering approach. The band images are clustered in the image space by computing their posterior class probability. Then, for each cluster, the band exhibiting the highest probability of belonging to it is selected as cluster exemplar. More particularly, the proposed method falls into information-maximization clustering methods, where the posterior class probability is modeled and the parameters of the models are derived by maximizing the information between the data and the unknown cluster labels. In this context, we propose a new image representation for hyperspectral images, based on the first and second order statistics of multiple image features. We refer to this representation as multiple-feature local statistical descriptors (MLSD). The descriptors are computed w.r.t. regular grids, and a special pixel selection procedure reduces the number of samples within each block of the grid. A kernel-based model that embeds the MLSD is then proposed for the posterior class probability. The model is finally optimized according to an information-maximization criterion. We conduct several experiments to determine the best parameters for the proposed approach and compare the latter with other state-of-the-art UBS methods. Quantitative evaluations show that, by employing our band selection method, higher performance in terms of classification accuracy and endmember extraction can be achieved in comparison with the state of the art.
Journal of the Optical Society of America, Oct 23, 2019
2022 30th European Signal Processing Conference (EUSIPCO)
Cet article presente une nouvelle architecture hybride basee sur l'encodage par vecteurs de F... more Cet article presente une nouvelle architecture hybride basee sur l'encodage par vecteurs de Fisher (VF) des sorties des couches convolutives d'un reseau de neurones. L'originalite de ce travail repose sur l'exploitation des statistiques d'ordre deux via le calcul des matrices de covariance locales. Considerant les proprietes intrinseques a la geometrie Riemannienne propre a l'espace des matrices de covariance, nous proposons d'utiliser la metrique log-euclidienne afin d'etendre le concept des VF pour l'encodage de matrices de covariance : les vecteurs de Fisher log-euclidiens (LE VF). L'architecture proposee est ensuite evaluee sur differentes bases de donnees de teledetection : la base UC Merced Land Use Land Cover, la base AID, ainsi que sur deux jeux de donnees Pleiades sur des forets de pins maritimes et de parcs ostreicoles.
In this paper, we introduce a new generative adversarial network (GAN) with dual image-color disc... more In this paper, we introduce a new generative adversarial network (GAN) with dual image-color discriminators, to predict Artificial-SAR colorized images from SAR ones (Sentinel-1). Based on the conventional architecture of GANs, we employ an additional color discriminator that evaluates the differences in brightness, contrast, and major colors between images, while the image discriminator compares texture and content. To achieve the required level of colorization in the generation process, we employ non-adversarial color loss dedicated for color comparison, unlike conventional approaches that use only L1 loss. Moreover, to overcome the vanishing gradient problem in deep architecture, and ensure the flow of low-level information inside network layers, we add residual connections to our generator that follows the general shape of U-Net. The performance of the proposed model was evaluated quantitatively as well as qualitatively with the SEN1-2 dataset. Results show that the proposed mod...
2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Mar 1, 2017
Many signal and image processing applications, including SAR polarimetry and texture analysis, re... more Many signal and image processing applications, including SAR polarimetry and texture analysis, require the classification of complex covariance matrices. The present paper introduces a geometric learning approach on the space of complex covariance matrices based on a new distribution called Riemannian Gaussian distribution. The proposed distribution has two parameters, the centre of massȲ and the dispersion parameter σ. After having derived its maximum likelihood estimator and its extension to mixture models, we propose an application to texture recognition on the VisTex database.
IEEE Transactions on Image Processing, 2014
This index covers all technical items-papers, correspondence, reviews, etc.-that appeared in this... more This index covers all technical items-papers, correspondence, reviews, etc.-that appeared in this periodical during 2014, and items from previous years that were commented upon or corrected in 2014. Departments and other items may also be covered if they have been judged to have archival value. The Author Index contains the primary entry for each item, listed under the first author's name. The primary entry includes the coauthors' names, the title of the paper or other item, and its location, specified by the publication abbreviation, year, month, and inclusive pagination. The Subject Index contains entries describing the item under all appropriate subject headings, plus the first author's name, the publication abbreviation, month, and year, and inclusive pages. Note that the item title is found only under the primary entry in the Author Index.
Journal of Applied Geophysics, 2018
For a geoscientist, the Relative Geologic Time (RGT) is an important tool to perform chronostrati... more For a geoscientist, the Relative Geologic Time (RGT) is an important tool to perform chronostratigraphic analysis. However, automatically estimate an RGT image from a seismic image can be a challenging task where we have to respect seismic features, the deposit orders and to deal efficiently with unconformities such as erosions, progradating systems, etc. To this end, approaches have been proposed formulating the estimation problem in a regularized convex optimization problem. However none of these fully address efficiently all issues. In this paper, we propose a new regularization term based on an asymmetric and adaptive weight function. The asymmetric behavior focuses on the ill-posed problem while the adaptive process is used to deal with unconformities by modulating the strength of the regularization if necessary. Moreover, to increase the robustness of the approach, we propose variants of the method in terms of ℓ 1-norm instead of ℓ 2-norm that corresponds to potentially too smooth solutions. For evaluating the relevance of our proposals, experimentations have been conducted on both synthetic and real seismic images.
Ce travail est parrainé par le Gdr Automatique « la dérivation non entière en isolation vibratoir... more Ce travail est parrainé par le Gdr Automatique « la dérivation non entière en isolation vibratoire » Résumé-Dans un contexte de synthèse de profil routier, cet article présente l'extension d'une méthode de synthèse monodimensionnelle à la génération d'un terrain fractal auto-similaire 2-D. L'objectif est de fournir un terrain dont la dimension fractale est isotrope en terme d'évolution spatiale. La méthode de synthèse s'appuie sur une décomposition de Cholesky. Nous montrons en outre le lien qu'il existe entre les matrices issues de la décomposition selon l'élongation longitudinale et transversale.
Physics in Medicine and Biology, 2006
Non-invasive methods for quantifying [(18)F]FDG uptake in tumours often require normalization to ... more Non-invasive methods for quantifying [(18)F]FDG uptake in tumours often require normalization to either body weight or body surface area (BSA), as a surrogate for [(18)F]FDG distribution volume (DV). Whereas three dimensions are involved in DV and weight (assuming that weight is proportional to volume), only two dimensions are obviously involved in BSA. However, a fractal geometry interpretation, related to an allometric scaling, suggests that the so-called 'body surface area' may stand for DV.
This paper presents a novel framework for visual content classification using jointly local mean ... more This paper presents a novel framework for visual content classification using jointly local mean vectors and covariance matrices of pixel level input features. We consider local mean and covariance as realizations of a bivariate Riemannian Gaussian density lying on a product of submanifolds. We first introduce the generalized Mahalanobis distance and then we propose a formal definition of our product-spaces Gaussian distribution on R m × SPD(m). This definition enables us to provide a mixture model from a mixture of a finite number of Riemannian Gaussian distributions to obtain a tractable descriptor. Mixture parameters are estimated from training data by exploiting an iterative Expectation-Maximization (EM) algorithm. Experiments in a texture classification task are conducted to evaluate this extended modeling on several color texture databases, namely popular Vistex, 167-Vistex and CUReT. These experiments show that our new mixture model competes with state-of-the-art on the experimented datasets.
arXiv (Cornell University), Jun 7, 2022
-Les méthodes de super résolution d'image visentà recréer une image haute résolutionà partir d'un... more -Les méthodes de super résolution d'image visentà recréer une image haute résolutionà partir d'une basse résolution. La famille d'approches basée sur les patchs a fait l'objet d'une attention et d'un développement considérable. La technique de minimum de l'erreur quadratique moyenne est une méthode de restauration d'images qui utilise un modèle gaussien de probabilités sur les patchs d'images. Cet article propose un algorithme d'apprentissage d'un modèle conjoint de mélange gaussien généralisé (GGMM)à partir des paires de patchsà basse résolution et des patchs correspondantsà haute résolution d'une image de référence.À partir de ce modèle GGMM, l'image haute résolution en utilisant la méthode MMSE. Nosévaluations numériques indiquent que la méthode MMSE-GGMM se comporte très bien par rapportà l'état de l'art.
Pattern Recognition, Oct 1, 2023
Lecture Notes in Computer Science, 2022
IEEE Transactions on Geoscience and Remote Sensing, Aug 1, 2016
In order to obtain accurate classification results of hyperspectral images, both the spectral and... more In order to obtain accurate classification results of hyperspectral images, both the spectral and spatial information should be fully exploited in the classification process. In this paper, we propose a novel method using independent component analysis (ICA) and edge-preserving filtering (EPF) via an ensemble strategy for the classification of hyperspectral data. First, several subsets are randomly selected from the original feature space. Second, ICA is used to extract spectrally independent components followed by an effective EPF method, to produce spatial features. Two strategies (i.e., parallel and concatenated) are presented to include the spatial features in the analysis. The spectralspatial features are then classified with a random forest (RF) or rotation forest (RoF) classifier. Experimental results on two real hyperspectral datasets demonstrate the effectiveness of the proposed methods. A sensitivity analysis of the new classifiers is also performed. Index Terms Classification, hyperspectral data, independent component analysis (ICA), edge preserving filter (EPF). I. INTRODUCTION During the past two decades, the development of hyperspectral sensors have resulted in great improvement for the image acquisition capabilities. Hyperspectral sensors are now able to provide images with both high spectral and spatial resolutions. Hence, hyperspectral data offers a unique opportunity to monitor the Earth surface. Thematic applications include environmental mapping and crop monitoring [1], [2]. Supervised classification is one of the most important problems in the remote sensing community. Given a set of training samples (i.e., pixel vectors for hyperspectral image), the aim of classification is to assign a unique Manuscript received ; revised. This study has been carried out with financial support from the French State, managed by the French National Research Agency (ANR) in the frame of the "Investments for the future" Programme IdEx Bordeaux-CPU (ANR-10-IDEX-03-02).