Ce mémoire de thèse porte sur l’élaboration de systèmes de reconnaissance d’image qui sont robustes à la variabilité géométrique. La compréhension d’une image est un problème difficile, de par le fait qu’elles sont des projections en deux dimensions d’objets 3D. Par ailleurs, des représentations qui doivent appartenir à la même catégorie, par exemple des objets de la même classe en classification, peuvent être visuellement très différentes. Notre but est de rendre ces systèmes robustes à la juste quantité de déformations, celle-ci étant automatiquement déterminée à partir des données. Nos deux contributions sont les suivantes. Nous montrons tout d’abord comment utiliser des exemples virtuels pour rendre les systèmes de classification d’images robustes et nous proposons ensuite une méthodologie pour apprendre des descripteurs de bas niveau robustes, pour la recherche d’image.Nous étudions tout d’abord les exemples virtuels, en tant que transformations de vrais exemples. En représentant une image en tant que sac de descripteurs transformés, nous montrons que l’augmentation de données, c’est-à-dire le fait de les considérer comme de nouveaux exemples iid, est la meilleure manière de les utiliser, pourvu qu’une étape de vote avec les descripteurs transformés soit opérée lors du test. Du fait que les transformations apportent différents niveaux d’information, peuvent être redondants, voire nuire à la performance, nous pro-posons un nouvel algorithme capable de sélectionner un petit nombre d’entre elles,en maximisant la justesse de classification. Nous montrons par ailleurs comment remplacer de vrais exemples par des virtuels, pour alléger les couts d’annotation.Nous rapportons de bons résultats sur des bancs d’essai de classification.Notre seconde contribution vise à améliorer les descripteurs de régions locales utilisés en recherche d’image, et en particulier nous proposons une alternative au populaire descripteur SIFT. Nous proposons un nouveau descripteur, appelé patch-CKN, appris sans supervision. Nous introduisons un nouvel ensemble de données liant les images et les imagettes, construit à partir de reconstruction3D automatique d’images récupérées sur Internet. Nous définissons une méthode pour tester précisément la performance des descripteurs locaux au niveau de l’imagette et de l’image. Notre approche dépasse SIFT et les autres approches à base d’architectures convolutionnelles sur notre banc d’essai, et d’autres couramment utilisés dans la littérature. / This dissertation focuses on designing image recognition systems which are robust to geometric variability. Image understanding is a difficult problem, as images are two-dimensional projections of 3D objects, and representations that must fall into the same category, for instance objects of the same class in classification can display significant differences. Our goal is to make systems robust to the right amount of deformations, this amount being automatically determined from data. Our contributions are twofolds. We show how to use virtual examples to enforce robustness in image classification systems and we propose a framework to learn robust low-level descriptors for image retrieval. We first focus on virtual examples, as transformation of real ones. One image generates a set of descriptors –one for each transformation– and we show that data augmentation, ie considering them all as iid samples, is the best performing method to use them, provided a voting stage with the transformed descriptors is conducted at test time. Because transformations have various levels of information, can be redundant, and can even be harmful to performance, we propose a new algorithm able to select a set of transformations, while maximizing classification accuracy. We show that a small amount of transformations is enough to considerably improve performance for this task. We also show how virtual examples can replace real ones for a reduced annotation cost. We report good performance on standard fine-grained classification datasets. In a second part, we aim at improving the local region descriptors used in image retrieval and in particular to propose an alternative to the popular SIFT descriptor. We propose new convolutional descriptors, called patch-CKN, which are learned without supervision. We introduce a linked patch- and image-retrieval dataset based on structure from motion of web-crawled images, and design a method to accurately test the performance of local descriptors at patch and image levels. Our approach outperforms both SIFT and all tested approaches with convolutional architectures on our patch and image benchmarks, as well as several styate-of-theart datasets.

The ascidian (subphylum Tunicata: class Ascidiacea) fauna along the southern African coast constitutes an important component of the sessile benthic reef fauna. Little is known of the species composition, biodiversity and distribution of ascidians on southern African intertidal and subtidal reefs. Past research on the ascidian fauna of South Africa was sporadic and limited, with only about one hundred and seventy species recorded during the last one hundred and twenty years. This is the first taxonomic study undertaken by a South African resident scientist. The study focused on six genera in four families (Euherdmaniidae, Polyclinidae, Pseudodistomidae and Didemnidae) belonging to the class Ascidiacea and suborder Aplousobranchia. Five species new to science, two Polyclinum, two Pseudodistoma and one Polysyncraton species are described. New additional information on the distribution of five species previously known to science, including one recorded for the first time along the South African coast, is presented. A literature review and comparison of the taxonomic important characteristics is made of all species known globally for the six genera. These six genera comprise twelve Euherdmania, fifteen Aplidiopsis, forty Polyclinum thirty-one Pseudodistoma, six Atriolum and sixty-eight Polysyncraton species.

No description available.

Doctorat en Sciences / info:eu-repo/semantics/nonPublished

Depuis 2007, une nouvelle génération de radar satellitaires imageurs plus performants permet de renouveler les approches cartographiques et ce, en particulier, sur les zones tropicales où l'acquisition des images optiques est difficile du fait de la présence quasi permanente de nuages. Ces radars opèrent en différentes longueurs d'onde (bande L, C, X), différentes polarisations et avec des résolutions spatiales variables (Très Haute Résolution à Moyenne Résolution Spatiale).Nos sites d'étude se situent en Guyane française et en Côte d'Ivoire en contexte climatique tropical humide. Sur la Guyane française de nombreuses études antérieures garantissent une connaissance géographique riche et détaillée, nos recherches porteront sur la méthodologie de traitement et d'analyse d'images radar multi-paramètres (longueur d'onde, polarisation, résolution) et de leur exploitation conjointe avec des images optiques très haute résolution spatiale en vue d'effectuer le suivi temporel des milieux rencontrés localement (forêt dense sur socle, plaine littorale, forêt de mangrove, zones humides…).Une fois ce bilan détaillé et argumenté, effectué sur la Guyane, les recherches se porteront sur des zones test en Côte d'Ivoire qui contrairement à la Guyane, ne disposent pas de connaissances géographiques de qualité mais où le climat tropical rend, comme en Guyane, l'usage des images radar incontournable. Le premier chapitre de ce travail décrit les milieux rencontrés sur nos zones d'étude ainsi que les données disponibles sur chacun d'entre eux. Le second chapitre porte sur la méthodologie de traitement et d'analyse d'images radar multi paramètres et de leur exploitation conjointe avec des images optiques très haute résolution spatiale. Deux approches d'évaluation des images y sont discutées. Une première approche, basée sur la photo-interprétation, s'appuie sur les documents cartographiques existants, les levés de terrain pour évaluer visuellement les images. Une seconde approche aborde l'évaluation grâce aux méthodes de classification supervisée pour lesquelles deux techniques ont été utilisées : la classification SVM et la classification orientée objet. Le troisième chapitre est consacré aux applications thématiques des images sur les zones test situées en Guyane française et en Côte d'Ivoire. Y sont abordées la cartographie de l'occupation du sol, le suivi des zones humides littorales et la dynamique côtière et toutes ces applications pouvant conduire à la mise à jour des documents cartographiques existants / Since 2007, the launch of fully polarimetric SAR sensors has significantly improved the potential of satellite SAR data for the vegetation cartography. These sensors allow to make a complete description of the polarization state of the backscattered wave whatever the polarization configuration of the incident wave. Polarimetric data are sensitive to the geometrical structure of the vegetation cover, bringing a significant contribution for vegetation cartography. Different fully polarimetric sensors have been launched since 2007: ComoSkyMed, RADARSAT-2, and PALSAR sensors allowing observations in X, C, and L bands, with spatial resolution ranging from 1 to 30 m. Pilote sites are French Guiana (French) and Ivory coast (West African country) in humid tropical zone. In French Guiana, many studies in the past years guarantee a rich and detailed geographical knowledge, our works concern the multi parameters radar image analysis method (wavelength, polarization, resolution) and their exploitation together with external mapping data in the aim of implementing the temporal survey of the locally encountered environment (dense forest, coastal plains, mangrove forests, wetlands,…). In fact, with 80% of forest cover, French Guyana needs an update of existing cartographic documents to quantify the evolution of his dynamic vegetation. Once this assessment made on the Guyana was detailed and argued, the researches will concern to test zones in Ivory Coast which contrary to the Guyana, do not have geographical quality knowledge but where the tropical climate makes the use of radar images necessary as in Guyana. The first chapter of this work describes the fields met on our areas of study as well as the data available on each of them. Second chapter focuses on the methodology of processing and analysis of multicriterias radar images and their joint exploitation with optical images of very high-spatial resolution. Then, two approaches of evaluation of the images are discussed. A first approach, based on photo-interpretation, relies on the existing cartographic documents and on the land surveys to estimate visually the images. A second approach stating the evaluation thanks to the supervised classification for which two techniques were used: the SVM classification and the object-oriented classification. The third chapter is dedicated to the thematic applications of images on the test areas situated in French Guyana and in Ivory Coast. The mapping of the land use, the follow-up (survey) of the littoral wetlands and the coastal dynamics, all these applications can lead to the update of the existing cartographic documents

M.Sc. / Alooideae (Asparagales–Asphodelaceae) currently comprises five genera, four of which are endemic to southern Africa. Despite their importance in commercial horticulture, evolutionary relationships among the genera are still incompletely understood, and the generic taxonomy remains unresolved. This study addresses generic delimitation in subfamily Alooideae using an expanded molecular dataset from three plastid regions (rbcLa, matK and trnH-psbA) and the first subunit of the nuclear ribosomal internal transcribed spacer (ITS1). Sequence data were analysed using maximum parsimony and bayesian statistics, and selected morphological traits were mapped onto the molecular phylogeny. Among the polytypic genera, only Astroloba and Gasteria were retrieved as monophyletic. Haworthia is shown to be polyphyletic and to comprise three main clades largely correllated with current subgeneric circumscriptions. The taxonomic implications of these findings are discussed.

M.Sc. (Botany) / The subtribe Phymasperminae forms one of the six earliest diverging lineages within the tribe Anthemideae (Asteraceae/ Compositae). It comprises three genera, Eumorphia DC. (6 spp.), Gymnopentzia Benth. (1 sp.) and Phymaspermum Less. (ca. 19 spp.), all endemic to southern Africa. Generic and species delimitations and relationships within the subtribe are, however, uncertain. Anatomical, morphological and molecular characters are therefore explored in an attempt to resolve taxonomic delimitations and relationships within the subtribe. Two fruit characters, namely the presence of a papillose surface and more than 10 ribs are confirmed as synapomorphies for the subtribe. Phymaspermum is the largest and most complex genus within the subtribe and has been identified as one of the priority groups for taxonomic revision in South Africa. It is distinguished from the other two genera within Phymasperminae by the unique stalked myxogenic trichomes on the fruit surfaces. This character is shown to be a synapomorphy for Phymaspermum although subsequently lost in two species where it is replaced by resin canals in the fruit ribs. The species can be distinguished by a combination of habit, leaf, involucral and fruit characters. A taxonomic revision of Phymaspermum is presented, including comprehensive descriptions, a key to the species, nomenclature, typification, diagnostic characters with illustrations and maps with accurate geographical distribution data. Four new species are described, namely P. aphyllum, P. comptonii, P. oppositifolium and P. trifidum. Five previously recognised species, namely P. bolusii, P. montanum, P. equisetiodes, P. pubescens and P. villosum have been reduced to synonomy.

Doctor Scientiae (Biodiversity and Conservation Biology) / The purpose of this thesis was to provide a revision and characterization of the species of nongeniculate coralline algae previously ascribed to the now defunct genus Porolithon; to provide a modern account of selected taxa from the genus Hydrolithon and descriptions of taxa found to conform the generic delimitation of Hydrolithon and to use a phenetic cluster analysis to determine the taxonomic relationships between the various taxa ascribed to the genera Polorithon and Hydrolithon. / South Africa

The alpha taxonomy of the southern African Synodontis is reviewed, based on the analysis of 48 characters. The previously descriptive characters of barbel branching and humeral process shape are compared by various quantitative methods. The variation in colour pattern is recorded with the aid of detailed line drawings. The definitions of head length, head width and humeral process length are altered slightly from their traditional descriptions to ensure greater consistency and precision in measurement. New characters include the size of the caudal fork, the number, shape and arrangement of the premaxillary teeth, and two measurements between the cranium and pectoral girdle. Both uni- and bivariate methods are used to record and compare the intraspecific variation of each character. Principal components analysis of 19 morphometric characters is used to assess the degree of similarity of nine allopatric populations of S. zambezensis. Ten species are recognized from the study area, two of which are described as new: S. nebulosus Peters 1852, S. zambezensis Peters 1852, S. nigromaculatus Boulenger 1905, S. njassae Keilhack 1908, S. woosnami Boulenger 1911, S. macrostigma Boulenger 1911, S. leopardinus Pellegrin 1914, S. thamalakanensis Fowler 1935, S. macrostoma sp.n. and S.vanderwaali sp.n. A key to their identification is provided. Characters are discussed in terms of their contribution to the identification of the southern African species and, where possible, suggestions made concerning the value of these characters to the taxonomy of the genus as a whole. The rejection of certain characters previously used in Synodontis keys is discussed and alternatives proposed. The state of southern African Synodontis taxonomy is assessed and recommendations for future research are given

Le régime de l’auto-entrepreneur connaît depuis sa création un succès considérable en termes d’adhésions. Mais son principe phare de simplification en fait sa propre faiblesse. D’un régime social et fiscal incitatif constituant une alternative au chômage, l’auto-entrepreneuriat suscite aussi la tentation de l’abus. A l’examen attentif de ses motivations et de la réalisation économique de son projet-entrepreneurial, la réalité de l’auto-entrepreneur émerge. A l’image d’entrepreneur autonome libre de ses choix se superposent les nécessités d’un travailleur, employeur de lui-même et de nul autre, employé de lui-même à défaut de l’être par un autre. Après avoir analysé les séductions et les travers du régime, l’étude propose de sécuriser l’auto-entrepreneuriat en explorant les potentialités du droit du travail à l’accueillir. L’observation des dispositifs similaires des voisins européens dresse une perspective plausible d’émergence d’un droit de l’auto-entrepreneuriat dès lors qu’il n’attente pas à celui des autres acteurs économiques, en premier lieu celui des salariés. / Self-employment scheme have been considerably successful since their introduction. However, its flagship principle of simplification is also its main weakness. While self-employment schemes provide social and fiscal alternatives to unemployment, they can also be easily abused. The reality of self-employment emerges from an assessment of a project’s motivating factors and of its economic implementation. The image of an autonomous entrepreneur with freedom of choices becomes superimposed with the needs of a worker, employing himself and nobody else, employed by himself and not by someone else. After having analyzed the appeals and faults of self-employment scheme, this study proposes more secure self-employment principles by exploring the potential of its inclusion in labour law. An observation of similar mechanisms in neighboring European countries suggests a plausible perspective for the emergence of stronger self-employment right, as long as this does not violate the rights of other economic actors, primarily that of salaried employees.The image of an autonomous entrepreneur with freedom of choices becomes superimposed with the needs of a worker, employing himself and nobody else, employed by himself and not by someone else.