Academic literature on the topic 'HYPER SPECTRAL IMAGE CLASSIFICATION'

A rule-based classification model was developed to derive land-cover information from a large set of hyper-temporal, multi-spectral satellite imagery encompassing the state of Arizona. The model uses Advanced Very High Resolution Radiometer (AVHRR) imagery and the 30-minute digital elevation model (DEM) from the EROS Data Center (EDC) Conterminous U.S. AVHRR Biweekly Composites. Sixty one images from 1990, 1991 and 1992 were analyzed using the Brown & Lowe (1973) Natural Vegetative Communities of Arizona map to identify temporal patterns of Normalized Difference Vegetation Index (NDVI) and thermal measurements for 13 land-cover classes. Fifteen characteristic layers were created to represent the spectral, thermal and temporal properties of the data set. These layers were inputs for the rule-based classification model. The model was run on three years of data, creating three single year land-cover maps. The modeling effort showed that NDVI, thermal and DEM characteristics are useful for discerning land-cover classes. The single year land-cover maps showed that the rule-based model could not detect land-cover change between years. The single year maps were combined to create a summary land-cover map. This map differs from the Brown and Lowe map in the shape, proportional size and spatial distribution of land-cover polygons. The rule-based model can discern more land-cover classes than spectral cluster classification. Ground observations and an aerial video was used to assess map accuracy. The same proportion of agreement was observed between the ground observations, the Brown and Lowe map, and the summary land-cover map. Agreement was higher between video and the summary map than between video and the Brown and Lowe map. With further refinements to the input data set, classification model rules and field accuracy assessment, higher levels of agreement can be expected. Overall results show that rule-based classification of hyper-temporal, multi-spectral satellite imagery is a desirable method for mapping global land-cover.

Recent advances in sensor technology have led to an increased availability of hyperspectral remote sensing images with high spectral and spatial resolutions. These images are composed by hundreds of contiguous spectral channels, covering a wide spectral range of frequencies, in which each pixel contains a highly detailed representation of the reflectance of the materials present on the ground, and a better characterization in terms of geometrical detail. The burst of informative content conveyed in the hyperspectral images permits an improved characterization of different land coverages. In spite of that, it increases significantly the complexity of the analysis, introducing a series of challenges that need to be addressed, such as the computational complexity and resources required. This dissertation aims at defining novel strategies for the analysis and classification of hyperspectral remote sensing images, placing the focal point on the investigation and optimisation techniques for the extraction and integration of spectral and spatial information. In the first part of the thesis, a thorough study on the analysis of the spectral information contained in the hyperspectral images is presented. Though, independent component analysis (ICA) has been widely used to address several tasks in the remote sensing field, such as feature reduction, spectral unmixing and classification, its employment in extracting class-discriminant information remains a research topic open to further investigation. To this extend, a profound study on the performances of different ICA algorithms is performed, highlighting their strengths and weaknesses in the hyperspectral image classification task. Based on this study, a novel approach for feature reduction is proposed, where the use of ICA is optimised for the extraction of class-specific information. In the second part of the thesis, the spatial information is exploited by employing operators from the mathematical morphology framework. Morphological operators, such as attribute profiles and their multi-channel and multi-attribute extensions, are proved to be effective in the modelling of the spatial information, dealing, however, with issues such as the high feature dimensionality, the high intrinsic information redundancy and the a-priori need for parameter tuning in filtering, which are still open. Addressing the first two issues, the reduced attribute profiles are introduced, in this thesis, as an optimised version of the morphological attribute profiles, with the property to compress all the meaningful geometrical information into a few features. Regarding the filter parameter tuning issue, an innovative strategy for automatic threshold selection is proposed. Inspired by the concept of granulometry, the proposed approach defines a novel granulometric characteristic function, which provides information on the image decomposition according to a given measure. The approach exploits the tree representation of an image, allowing us to avoid additional filtering steps prior to the threshold selection, making the process computationally effective. The outcome of this dissertation advances the state-of-the-art by proposing novel methodologies for accurate hyperspectral image classification, where the results obtained by extensive experimentation on various real hyperspectral data sets confirmed their effectiveness. Concluding the thesis, insightful and concrete remarks to the aforementioned issues are discussed.

Recent advances in sensor technology have led to an increased availability of hyperspectral remote sensing images with high spectral and spatial resolutions. These images are composed by hundreds of contiguous spectral channels, covering a wide spectral range of frequencies, in which each pixel contains a highly detailed representation of the reflectance of the materials present on the ground, and a better characterization in terms of geometrical detail. The burst of informative content conveyed in the hyperspectral images permits an improved characterization of different land coverages. In spite of that, it increases significantly the complexity of the analysis, introducing a series of challenges that need to be addressed, such as the computational complexity and resources required. This dissertation aims at defining novel strategies for the analysis and classification of hyperspectral remote sensing images, placing the focal point on the investigation and optimisation techniques for the extraction and integration of spectral and spatial information. In the first part of the thesis, a thorough study on the analysis of the spectral information contained in the hyperspectral images is presented. Though, independent component analysis (ICA) has been widely used to address several tasks in the remote sensing field, such as feature reduction, spectral unmixing and classification, its employment in extracting class-discriminant information remains a research topic open to further investigation. To this extend, a profound study on the performances of different ICA algorithms is performed, highlighting their strengths and weaknesses in the hyperspectral image classification task. Based on this study, a novel approach for feature reduction is proposed, where the use of ICA is optimised for the extraction of class-specific information. In the second part of the thesis, the spatial information is exploited by employing operators from the mathematical morphology framework. Morphological operators, such as attribute profiles and their multi-channel and multi-attribute extensions, are proved to be effective in the modelling of the spatial information, dealing, however, with issues such as the high feature dimensionality, the high intrinsic information redundancy and the a-priori need for parameter tuning in filtering, which are still open. Addressing the first two issues, the reduced attribute profiles are introduced, in this thesis, as an optimised version of the morphological attribute profiles, with the property to compress all the meaningful geometrical information into a few features. Regarding the filter parameter tuning issue, an innovative strategy for automatic threshold selection is proposed. Inspired by the concept of granulometry, the proposed approach defines a novel granulometric characteristic function, which provides information on the image decomposition according to a given measure. The approach exploits the tree representation of an image, allowing us to avoid additional filtering steps prior to the threshold selection, making the process computationally effective. The outcome of this dissertation advances the state-of-the-art by proposing novel methodologies for accurate hyperspectral image classification, where the results obtained by extensive experimentation on various real hyperspectral data sets confirmed their effectiveness. Concluding the thesis, insightful and concrete remarks to the aforementioned issues are discussed.

Hyperspectral remote sensing image data, such as that recorded by AVIRIS with 224 spectral bands, provides rich information on ground cover types. However, it presents new problems in machine assisted interpretation, mainly in long processing times and the difficulties of class training due to the low ratio of number of training samples to the number of bands. This thesis investigates feasible and efficient feature reduction and image classification techniques which are appropriate for hyperspectral image data. The study is reported in three parts. The first concerns a deterministic approach for hyperspectral data interpretation. Multigroup and multiple threshold spectral coding procedures, and associated techniques for spectral matching and classification, are proposed and tested. By coding on subgroups of bands using one or three thresholds, spectral searching and matching becomes simple, fast and free of the need for radiometric correction. Modifications of existing statistical techniques are proposed in the second part of the investigation A block-based maximum likelihood classification technique is developed. Several subgroups are formed from the complete set of spectral bands in the data, based on the properties of global correlation among the bands. Subgroups which are poorly correlated with each other are treated independently using conventional maximum likelihood classification. Experimental results demonstrate that, when using appropriate subgroup sizes, the new method provides a compromise among classification accuracy, processing time and available training pixels. Furthermore, a segmented, and possibly multi-layer, principal components transformation is proposed as a possible feature reduction technique prior to classification, and for effective colour display. The transformation is performed efficiently on each of the highly correlated subgroups of bands independently. Selected features from each transformed subgroup can be then transformed again to achieve a satisfactory data reduction ratio and to generate the three most significant components for colour display. Classification accuracy is improved and high quality colour image display is achieved in experiments using two AVIRIS data sets.

Thesis (M.S. in Information Technology Management)--Naval Postgraduate School, March 2005.<br>Thesis Advisor(s): Richard C. Olsen. Includes bibliographical references (p. 95-97). Also available online.

The recent advances in aerial- and satellite-based hyperspectral imaging sensor technologies have led to an increased availability of Earth's images with high spatial and spectral resolution, which opened the door to a large range of important applications. Hyperspectral imaging records detailed spectrum of the received light in each spatial position in the image, in which each pixel contains a highly detailed representation of the reflectance of the materials present on the ground, and a better characterization in terms of geometrical details. Since different substances exhibit different spectral signatures, the abundance of informative content conveyed in the hyperspectral images permits an improved characterization of different land coverage. Therefore, hyperspectral imaging emerged as a well-suited technology for accurate image classi fication in remote sensing. In spite of that, a signi ficantly increased complexity of the analysis introduces a series of challenges that need to be addressed on a serious note. In order to fully exploit the potential offered by these sensors, there is a need to develop accurate and effective models for spectral-spatial analysis of the recorded data. This thesis aims at presenting novel strategies for the analysis and classifi cation of hyperspectral remote sensing images, placing the focal point on the investigation on deep networks for the extraction and integration of spectral and spatial information. Deep learning has demonstrated cutting-edge performances in computer vision, particularly in object recognition and classi cation. It has also been successfully adopted in hyperspectral remote sensing domain as well. However, it is a very challenging task to fully utilize the massive potential of deep models in hyperspectral remote sensing applications since the number of training samples is limited which limits the representation capability of a deep model. Furthermore, the existing architectures of deep models need to be further investigated and modifi ed accordingly to better complement the joint use of spectral and spatial contents of hyperspectral images. In this thesis, we propose three different deep learning-based models to effectively represent spectral-spatial characteristics of hyperspectral data in the interest of classifi cation of remote sensing images. Our first proposed model focuses on integrating CRF and CNN into an end-to-end learning framework for classifying images. Our main contribution in this model is the introduction of a deep CRF in which the CRF parameters are computed using CNN and further optimized by adopting piecewise training. Furthermore, we address the problem of over fitting by employing data augmentation techniques and increased the size of the training samples for training deep networks. Our proposed 3DCNN-CRF model can be trained to fully exploit the usefulness of CRF in the context of classi fication by integrating it completely inside of a deep model. Considering that the separation of constituent materials and their abundances provide detailed analysis of the data, our second algorithm investigates the potential of using unmixing results in deep models to classify images. We extend an existing region based structure preserving non-negative matrix factorization method to estimate groups of spectral bands with the goal to capture subtle spectral-spatial distribution from the image. We subsequently use these important unmixing results as input to generate superpixels, which are further represented by kernel density estimated probability distribution function. Finally, these abundance information-guided superpixels are directly supplied into a deep model in which the inference is implicitly formulated as a recurrent neural network to perform the eventual classifi cation. Finally, we perform a detailed investigation on the possibilities of adopting generative adversarial models into hyperspectral image classifi cation. We present a GAN-based spectral-spatial method that primarily focuses on signifi cantly improving the multiclass classi cation ability of the discriminator of GAN models. In this context, we propose to adopt the triplet constraint property and extend it to build a useful feature embedding for remote sensing images for use in classi cation. Furthermore, our proposed Triplet- 3D-GAN model also includes feedback from discriminator's intermediate features to improve the quality of the generator's sample generation process.<br>Thesis (PhD Doctorate)<br>Doctor of Philosophy (PhD)<br>School of Info & Comm Tech<br>Science, Environment, Engineering and Technology<br>Full Text

Le dimensionnement de capteur est un enjeu majeur pour le domaine de la détection d'aéronefs. Dans cette optique, il est nécessaire de simuler ces capteurs via des modèles et un nombre conséquent d'images spectrales d'aéronefs. L'obtention de ces images via des campagnes aériennes de mesure est toutefois onéreuse et difficile. Une simulation de ces données s'impose donc. Afin de répondre à ces besoins, des algorithmes d'illumination globale à haute dimension spectrale sont utilisés. Dans ces conditions, ces algorithmes posent des problèmes de consommation mémoire et de temps de calcul. Le projet de recherche de cette thèse s'inscrit dans le cadre de ces problématiques.Dans un premier temps, nous nous sommes focalisés sur l'algorithme du Path Tracing et la parallélisation GPUpour le rendu d'images spectrales. Nous avons d'abord analysé les problèmes de ce type de rendu sur GPU.Nous avons ensuite proposé une nouvelle méthode et un schéma de parallélisation spectral qui permettent de réduire significativement la consommation mémoire et les temps de calcul.Dans un second temps, nous avons cherché à réduire la charge de calcul spectrale de la simulation. À cet égard, nous avons proposé de généraliser le rendu spectral stochastique d'image dans l'espace CIE XYZ en rendu d'image spectrale stochastique. Cette méthode permet de rendre directement et de manière plus précise et rapide les canaux d'un capteur en diminuant la dimension spectrale de la simulation. Pour conclure, les travaux de cette thèse permettent de simuler de manière précise des images multi, hyper et ultra spectrales. Le temps interactif peut être atteint dans notre cas en multi et hyper spectrale<br>Sensor dimensioning is a major issue for the aircraft detection field. In this vein, it is appropriate to simulate these sensorsvia models and a consequent set of spectral images. The acquisition of these images via an airborne measure campaign is unfortunately costly and difficult. A robust and fast simulation of these data is hence very appealing.In order to answer these needs, global illumination methods in high spectral dimension are used. In these circumstances,these methods raise serious issues in term of memory consumption and of computing time. Our research project focuses on these problematics.In the first instance, we have focused on the Path Tracing method and its GPU parallelization for the spectral image rendering. We have investigated at first the issues of this kind of rendering on the GPU. Then we have proposed a new method and an efficient spectral parallelization pattern which allows us to reduce significantly the memory consumption and thecomputing time.In the second phase, we have investigated how to reduce the spectral computational load of the simulation. Inthat sense, we have proposed to generalize the stochastic spectral rendering of color (XYZ) image to the stochastic spectral image rendering. This new method renders directly the channels of a sensor which allows us to reduce the memory andthe computing requirements by reducing the spectral computational load of the simulation.To sum up, the works of this thesis allows us to simulate accurately multi, hyper and ultra spectral images. The interactive time can be achieved in our case in multi and hyper spectral resolution

La problèmatique dont nous nous occupons dans cette thèse est la classification automatique d'images bidimensionnelles, ainsi que la détection d'objets génériques dans des images. Les avancées de ce champ de recherche contribuent à l'élaboration de systèmes intelligents, tels que des robots autonomes et la création d'un web sémantique. Dans ce contexte, la conception de représentations d'images et de classificateurs appropriés constituent des problèmes ambitieux. Notre travail de recherche fournit des solutions à ces deux problèmes, que sont la représentation et la classification d'images. Afin de générer notre représentation d'image, nous extrayons des attributs visuels de l'image et construisons une structure de graphe basée sur les propriétés liées au relations de proximités entre les points d'intérêt associés. Nous montrons que certaines propriétés spectrales de ces graphes constituent de bons invariants aux classes de transformations géométriques rigides. Notre représentation d'image est basée sur ces propriétés. Les résultats expérimentaux démontrent que cette représentation constitue une amélioration par rapport à d'autres représentations similaires, mais qui n'intègrent pas les informations liées à l'organisation spatiale des points d'intérêt. Cependant, un inconvénient de cette méthode est qu'elle fait appel à une quantification (avec pertes) de l'espace des attributs visuels afin d'être combinée avec un classificateur Support Vecteur Machine (SVM) efficace. Nous résolvons ce problème en créant un nouveau classificateur, basé sur la distance au plus proche voisin, et qui permet la classification d'objets assimilés à des ensembles de points. La linéarité de ce classificateur nous permet également de faire de la détection d'objet, en plus de la classification d'images. Une autre propriété intéressante de ce classificateur est sa capacité à combiner différents types d'attributs visuels de manière optimale. Nous utilisons cette propriété pour formuler le problème de classification de graphes de manière différente. Les expériences, menées sur une grande variété de jeux de données, montrent les bénéfices quantitatifs de notre approche.

El análisis de imágenes ha impulsado muchos descubrimientos en la ciencia actual. Esta tesis se centra en el análisis de imágenes remotas para inspección aérea, exactamente en el problema de segmentación y clasificación de acuerdo al uso del suelo. Desde el nacimiento de los sensores hiperespectrales su uso ha sido vital para esta tarea ya que facilitan y mejoran sustancialmente el resultado. Sin embargo el uso de imágenes hiperespectrales entraña, entre otros, problemas de dimensionalidad y de interacción con los expertos. Proponemos mejoras que ayuden a paliar estos inconvenientes y hagan el problema mas eficiente.<br>Lately image analysis have aided many discoveries in research. This thesis focusses on the analysis of remote sensed images for aerial inspection. It tackles the problem of segmentation and classification according to land usage. In this field, the use of hyperspectral images has been the trend followed since the emergence of hyperspectral sensors. This type of images improves the performance of the task but raises some issues. Two of those issues are the dimensionality and the interaction with experts. We propose enhancements overcome them. Efficiency and economic reasons encouraged to start this work. The enhancements introduced in this work allow to tackle segmentation and classification of this type of images using less data, thus increasing the efficiency and enabling the design task specific sensors which are cheaper. Also, our enhacements allow to perform the same task with less expert collaboration which also decreases the costs and accelerates the process.