Imagerie hyperspectrale

Résumé

vignette|Projection bi-dimensionnelle d'une image hyperspectrale d'une région de la Terre prise depuis l'espace. vignette|Image hyperspectrale de plusieurs pierres permettant d'identifier les éléments qui les composent. vignette|L'imagerie hyperspectrale comparée à l'imagerie spectrale. vignette|Les différentes techniques d'acquisition d'une image hyperspectrale. L'imagerie hyperspectrale ou spectro-imagerie est une technologie permettant d'obtenir l'image d'une scène dans un grand nombre (généralement plus d'une centaine) de bandes spectrales à la fois étroites et contigües. Elle permet en premier lieu de déterminer la composition chimique d'une surface photographiée et en fournit des indications sur la concentration de ceux-ci et les propriétés physiques. Elle est également utilisée dans différents processus industriels principalement pour la détermination de la qualité des produits. L'imagerie hyperspectrale est utilisée dans les prises d'images réalisées à bord d'avion depuis les

Source officielle

https://fr.wikipedia.org/wiki/Imagerie_hyperspectrale

À propos de ce résultat

Cette page est générée automatiquement et peut contenir des informations qui ne sont pas correctes, complètes, à jour ou pertinentes par rapport à votre recherche. Il en va de même pour toutes les autres pages de ce site. Veillez à vérifier les informations auprès des sources officielles de l'EPFL.

Publications associées (74)

Chargement

Afficher plus

Publications associées

Chargement

Personnes associées (9)

Personnes associées

Chargement

Unités associées (7)

Unités associées

Chargement

Concepts associés (16)

Concepts associés

Chargement

Cours associés (23)

Cours associés

Chargement

Séances de cours associées (34)

Séances de cours associées

Chargement

Publications associées (74)

Multispectral imaging

EPFL1997

Chargement

Light microscopy is a tool of paramount importance for biologists and has been constantly improved for the past four centuries. Despite many recent developments, microscopy techniques still require improvement, especially to reach better temporal and spectral resolutions. In particular, many high-end microscopes favor mostly spatial resolution, at the expense of the latter two types of resolution. In this thesis, we present methods based on the use of active illumination and computational algorithms to increase temporal and spectral resolutions of microscopes. Our methods aim to provide users with the flexibility to chose, within a single instrument, which type of resolution is to be favored based on the application at hand. More generally, our approach has fundamental implications on the signal sensing procedure, allowing, for example, to mitigate temporal aliasing in sequences of images. Our first method performs temporal super-resolution imaging of monochrome scenes using a hue-encoded shutter. By making use of an active multi-spectral illumination, temporal information is encoded in the hue of the acquisitions. We characterize the method showing a resolution improvement of 2.8 and an increase of frame-rate of a factor 3. We demonstrate the applicability of our method to bright-field transmission microscopy by applying the method to the beating heart of a zebrafish.We then extend this method to fluorescence microscopy. We add a temporal regularization term to make the method robust to fluorescent labelings inhomogeneities. We present an application of the method to the beating heart of a zebrafish that emits fluorescent light of two different colors. Implementing our method within a light-sheet microscope allows us to reconstruct 3D+time videos of the beating heart at twice the acquisition frame-rate. Our second method offers a way to perform temporal generalized sampling by computing simultaneous inner products with the sampled signal. Similarly to the first method, we take advantage of working with multiple illumination hues to compute as many simultaneous inner products, which we retrieve via an unmixing procedure. We use equivalent basic and dual B-splines representations to ensure having finite-length and positive pre-filters, as well as finite-support reconstruction functions. We show applications of our method to a fast rotating target, as well as to the beating heart of a zebrafish, both in transmission and fluorescence microscopy. Finally, we introduce a method to perform spectral imaging of repeating processes, such as the beating heart. The method sequentially acquires multiple movies with various filters, performs temporal registration of all movies and reconstructs a spectral movie through solving of a spectral unmixing problem, pixel by pixel, at each time point. We characterize the method and show a median error of approximately 10%, by comparing reconstructions on a static sample from our method with measurements obtained with a spectrometer. We then perform validation by comparing static reconstructions with dynamic ones of the same sample. We demonstrate the potential of the method to microscopy by performing spectral imaging of the beating heart of a zebrafish. Taken together, these methods offer a versatile toolbox to improve the temporal or spectral resolution in both bright field and fluorescence microscopy, which we foresee could be directly implemented in a number of specialized instruments.

EPFL2020

Chargement

Geometry-Aware Deep Recurrent Neural Networks for Hyperspectral Image Classification

Siyuan Hao, Mathieu Salzmann, Wei Wang

Variants of deep networks have been widely used for hyperspectral image (HSI)-classification tasks. Among them, in recent years, recurrent neural networks (RNNs) have attracted considerable attention in the remote sensing community. However, complex geometries cannot be learned easily by the traditional recurrent units [e.g., long short-term memory (LSTM) and gated recurrent unit (GRU)]. In this article, we propose a geometry-aware deep recurrent neural network (Geo-DRNN) for HSI classification. We build this network upon two modules: a U-shaped network (U-Net) and RNNs. We first input the original HSI patches to the U-Net, which can be trained with very few images and obtain a preliminary classification result. We then add RNNs on the top of the U-Net so as to mimic the human brain to refine continuously the output-classification map. However, instead of using the traditional dot product in each gate of the RNNs, we introduce a Net-Gated GRU that increases the nonlinear representation power. Finally, we use a pretrained ResNet as a regularizer to improve further the ability of the proposed network to describe complex geometries. To this end, we construct a geometry-aware ResNet loss, which leverages the pretrained ResNet's knowledge about the different structures in the real world. Our experimental results on real HSIs and road topology images demonstrate that our approach outperforms the state-of-the-art classification methods and can learn complex geometries.

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC2021

Chargement

Afficher plus

EPFL2020