Cette thèse est consacrée à la détermination de l’émissivité des surfaces terrestres (LSE) à partir de données dans les domaines des micro-ondes et de l’infrarouge thermique. (1) Ce travail a permis de fournir une méthode de détermination du LSE à partir des données AMSR-E (Advanced Microwave Scanning Radiometer-Earth Observing System) et de développer un modèle de transfert radiatif sol-atmosphère (SARTM) utilisé pour simuler les températures de brillance au niveau du capteur. Le modèle SARTM est construit à partir du modèle MonoRTM (MONOchromatic Radiative Transfer Model) et du modèle AIEM (Advanced Integral Equation Model). Dans cette étude les émissivités micro-ondes sur toute la Chine pour l’année 2006 ont été estimées. (2) Cette thèse présente également les améliorations apportées à un algorithme de détermination des émissivités à partir du capteur SEVIRI (Spinning Enhanced Visible and Infrared Imager) à bord du satellite MSG-2. Cet algorithme perfectionné est appliqué à plusieurs images MSG-2/SEVIRI sur une région d’étude de la péninsule ibérique. Il est démontré sur des cas détaillés que les améliorations portées sur la méthode originale de détermination du LSE et de la température de surface étaient réelles et cohérentes. / This thesis focused on the retrievals of Land Surface Emissivity (LSE) from microwave data and thermal infrared data. (1) This thesis provides a method to retrieval LSE from the AMSR-E (Advanced Microwave Scanning Radiometer-Earth Observing System) and develops a Soil-Atmosphere Radiative Transfer Model (SARTM) to simulated brightness temperatures at satellite level. SARTM model is built from MonoRTM (MONOchromaticRadiative Transfer Model) and from AIEM (Advanced Integral Equation Model) models. In this study, the LSEs over whole China of year 2006 are estimated. (2) This thesis also presents an improved algorithm to retrieve LSE from SEVIRI (Spinning Enhanced Visible and Infrared Imager) data onboardthe MSG-2 satellite. Finally, this improved algorithm is applied to several MSG-2/SEVIRI datasets over a study area withinthe Iberian Peninsula region. It is demonstrated with some detailed cases that these improvements on the original LSE/land surface temperature (LST) retrieval methods are effective and reasonable.

Soil attributes directly influence on its surface temperature. Although there are several studies using soil spectra obtained from satellites, soil evaluation through Land Surface Temperature (LST) is still scarce. The broad availability of satellite thermal data and the development of algorithms to retrieve LST facilitated its use in soil studies. The objective of this study was to evaluate soil LST variations due to its composition and verify the potential of using LST on soil attributes quantification, also integrated with reflectance spectra and elevation data. The study area (198 ha) is located in Sao Paulo state, Brazil, and had plowed bare soil during the satellite image acquisition date. Soil samples were collected in a regular grid of 100 x 100 m (depths: 0-0.2 m and 0.8- 1.0 m); soil granulometry, organic matter (OM) and iron oxides were determined by wet chemistry analysis. In this study, an image of Landsat 5 was used for extracting LST using the inversion of Planck\'s function in band 6 (10,400 - 12,500 nm), and land surface emissivity was estimated using Normalized Difference Vegetation Index threshold method. Reflectance values were extracted from bands 1, 2, 3, 4, 5 and 7. Models for soil attributes quantification were performed using Linear Regression (LR), with samples from 62 auger points distributed in 14 toposequences. Simple LR was applied for generating prediction models based on LST and on elevation data (extracted from a Digital Elevation Model). Multiple LR was applied in order to generate prediction models using atmospherically corrected spectral reflectance from Visible, Near-Infrared and Shortwave infrared (Vis-NIR-SWIR) bands as predictors, and also for the prediction of soil attributes using simultaneously Vis-NIR-SWIR, LST and elevation data, and only significant variables identified by T-tests were used. Predictive performance of models was assessed based on adjusted coefficient of determination (R2adj), Root Mean Squared Error (RMSE, g kg-1) and Ratio of Performance to Interquartile Range (RPIQ) obtained in validation. Ordinary kriging was also performed and the resulted interpolated surfaces were compared to the maps obtained from the best LR model. There was significant correlation between soil attributes and reflectance, LST and elevation data, and soils with clay texture were differentiated from sandy soils based on LST mean values. For all soil attributes, models using only elevation presented the worst performance; models using only LST, moderate performance; and using Vis-NIR-SWIR bands, good predictive performance. For clay, the best model obtained had bands 4-7, LST and elevation as predictors; for sand and iron oxides, the best model had bands 4-7 and LST; for OM, band 4, band 7 and LST. The use of LST for estimating soil attributes increases the predictive performance of multiple LR models when associated with other variables obtained through remote sensing, particularly surface reflectance data, improving the validation of models reaching high R2adj, high RPIQ and low RMSE values. Maps for sand, OM and iron oxides obtained through ordinary kriging outperformed those obtained for the same attributes using LR models based on RS co-variables, and for clay, both approaches reached the same accuracy level. Mapping of soil clay, sand, OM and iron oxides contents through multiple LR models using Landsat 5 products is a simple and easy to reproduce technique, appropriate for soil attributes mapping in bare soil agricultural areas. / Os atributos do solo influenciam diretamente na sua temperatura de superfície. Apesar de existir vários estudos utilizando espectros de solos obtidos de satélite, a avaliação do solo por meio da Temperatura de Superfície Terrestre (em inglês Land Surface Temperature, LST) ainda é escassa. A ampla disponibilidade de dados termais de satélite e o desenvolvimento de algoritmos para derivar a LST facilitou o seu uso em estudos de solos. O objetivo desse trabalho foi avaliar variações da LST do solo devidas à sua composição e verificar o potencial de uso da LST na quantificação de atributos do solo, também integrada com dados de espectros de reflectância e elevação. A área de estudo (198 ha) está localizada no estado de São Paulo, Brasil, e estava com solo exposto e arado na data de aquisição da imagem de satélite. Amostras de solo foram coletadas em um grid regular de 100 x 100 m (profundidades: 0.02 m e 0.8-1.0 m); a granulometria do solo, matéria orgânica (MO) e óxidos de ferro foram determinados via análises físicas e químicas laboratoriais. Neste estudo, uma imagem do Landsat 5 foi utilizada para extrair a temperatura de superfície usando a inversão da função da Lei de Planck na banda 6 (10.400 - 12.500 nm), e a emissividade de superfície foi estimada utilizando o método do limiar do Índice de Vegetação da Diferença Normalizada. Valores de reflectância das bandas 1, 2, 3, 4, 5 e 7 foram extraídos. Modelos para quantificação de atributos do solo foram feitos usando Regressão Linear (RL), com amostras de 62 pontos de tradagem distribuídos em 14 topossequências. A RL simples foi aplicada para gerar modelos de predição baseados na LST e também na elevação (extraída de um modelo digital de elevação). A RL múltipla foi aplicada para gerar modelos de predição usando os espectros de reflectância com correção atmosférica das bandas do Visível, Infravermelho próximo e Infravermelho de ondas curtas (Vis-NIR-SWIR) como preditores; também foi aplicada para predição de atributos do solo usando simultaneamente dados do Vis-NIR-SWIR, LST e elevação, e apenas variáveis significativas identificadas por teste T foram usadas. A performance preditiva dos modelos foi avaliada baseada no coeficiente de determinação ajustado (R2adj), raiz do erro quadrático médio (RMSE, g kg-1) e razão de desempenho do intervalo interquartil (RPIQ) obtidos na validação. A krigagem ordinária também foi feita e as superfícies interpoladas resultantes foram comparadas com o melhor modelo de RL. Houve correlação significativa entre os atributos do solo e dados de reflectância, LST e elevação, e solos com textura argilosa foram diferenciados de solos arenosos com base em valores médios de LST. Para todos os atributos do solo, os modelos usando apenas elevação apresentaram a pior performance, modelos usando somente LST, performance moderada, e usando as bandas do Vis-NIR-SWIR, boa performance preditiva. Para argila, o melhor modelo obtido teve as bandas 4-7, LST e elevação como preditores; para areia e óxidos de ferro, o melhor modelo teve as bandas 4-7 e LST; para MO, banda 4, banda 7 e LST. O uso da LST para estimar atributos do solo aumenta a performance preditiva de modelos de RL múltipla quando associada a outras variáveis obtidas via sensoriamento remoto (SR), particularmente dados de reflectância de superfície, melhorando a validação dos modelos atingindo altos valores de R2adj e RPIQ e baixos valores de RMSE. Os mapas para areia, MO e óxidos de ferro obtidos via krigagem ordinária superaram aqueles obtidos para os mesmos atributos usando modelos de RL baseados em co-variáveis obtidas via SR, e para argila, ambas abordagens atingiram o mesmo nível de acurácia. O mapeamento dos conteúdos de argila, areia, matéria orgânica e óxidos de ferro do solo via modelos de RL múltipla utilizando produtos do Landsat 5 é uma técnica simples e fácil de reproduzir, apropriada para o mapeamento de atributos do solo em áreas de agricultura com solo exposto.

熱紅外波段在遙感探測中佔有相當重要的地位，因其不受日夜條件限制，且因溫度變化時常具有與自然環境相關的特殊意義，使熱紅外影像可應用於測量、環境監控、都市開發、災害防治等領域。 在判釋遙測影像之前，通常必先確定各波段影像的幾何性質一致，若想將熱紅外影像與可見光影像套疊，須先率定蒐集熱影像之儀器，使影像受儀器本身的影響減到最低。本研究以FLIR-T360紅外線熱像儀為研究對象，探討熱像儀的成像特性，且嘗試率定與改正蒐集之熱影像。 率定熱像儀的實驗可分為幾何與輻射兩方面，幾何方面使用改良型的實地率定法，以求取熱像儀的內方位參數；輻射方面則使用實地調查法，求得控制點溫度，利用拍攝控制點蒐集多筆資料，擬合出輻射改正模型；本文亦展示熱影像幾何以及輻射改正後的成果。 / Thermal infrared data is important when conducting remote sensing investigation, for it could be acquired both in day and night. The change of temperature has characteristic significance of representing. So the thermal infrared images are used not only in the domain of surveying, but also in the environment monitoring, the urban development, and the disaster prevention. Before interpreting the remote sensing data, one would make sure that each image of bands has similar image geometry. Calibrating such geometry could prove that the effect from the lens distortion had been minimized. In such case, calibrated thermal images are necessary to guarantee that the image coordinates will correspond with the space coordinates as other bands. A thermal sensor, FLIR-T360 has been calibrated in this research. Two aspects of calibration executed are geometric and radiometric. A conventional calibrated template has been improved for using in the geometric aspect. The thermal sensor’s interior orientation elements were then found by using a field method. In the radiometric aspect, in situ method has been employed to determine temperatures of the chosen control points. The result of correction in geometric and radiometric aspect are also shown and discussed in this study.

The use of remote sensing technology has made it possible for the non-contact measurement of soil moisture content (SMC). Many remote sensing techniques can be used such as microwave sensors, electromagnetic waves sensors, capacitance, and thermal infrared sensors. Some of those techniques are constrained by their high fabrication cost, operation cost, size, or complexity. In this study, a thermal infrared technique was used to predict soil moisture content with the aid of using weather meteorological variables. <p>The measured variables in the experiment were soil moisture content (%SMC), soil surface temperature (Ts) measured using thermocouples, air temperature (Ta), relative humidity (RH), solar radiation (SR), and wind speed (WS). The experiment was carried out for a total of 12 soil samples of two soil types (clay/sand) and two compaction levels (compacted/non-compacted). After data analysis, calibration models relating soil moisture content (SMC) to differential temperature (Td), relative humidity (RH), solar radiation (SR), and wind speed (WS) were generated using stepwise multiple linear regression of the calibration data set. The performance of the models was evaluated using validation data. Four mathematical models of predicting soil moisture content were generated for each soil type and configuration using the calibration data set. Among the four models, the best model for each soil type and configuration was determined by comparing root mean of squared errors of calibration (RMSEC) and root mean of squared errors of validation (RMSEV) values. Furthermore, a calibration model for the thermal infrared sensor was developed to determine the corrected soil surface temperature as measured by the sensor (Tir) instead of using the thermocouples. The performance of the thermal infrared sensor to predict soil moisture content was then tested for sand compacted and sand non-compacted soils and compared to the predictive performance of the thermocouples. This was achieved by using the measured soil surface temperature by the sensor (Tir), instead of the measured soil surface temperature using the thermocouples to determine the soil-minus-air temperature (Td). The sensor showed comparable prediction performance, relative to thermocouples. <p>Overall, the models developed in this study showed high prediction performance when tested with the validation data set. The best models to predict SMC for compacted clay soil, non-compacted clay soil, and compacted sandy soil were three-variable models containing three predictive variables; Td, RH, and SR. On the other hand, the best model to predict SMC for compacted sandy soil was a two-variable model containing Td, and RH. The results showed that the prediction performance of models for predicting SMC for the sandy soils was superior to those of clay soils.

The use of remote sensing technology has made it possible for the non-contact measurement of soil moisture content (SMC). Many remote sensing techniques can be used such as microwave sensors, electromagnetic waves sensors, capacitance, and thermal infrared sensors. Some of those techniques are constrained by their high fabrication cost, operation cost, size, or complexity. In this study, a thermal infrared technique was used to predict soil moisture content with the aid of using weather meteorological variables. <p>The measured variables in the experiment were soil moisture content (%SMC), soil surface temperature (Ts) measured using thermocouples, air temperature (Ta), relative humidity (RH), solar radiation (SR), and wind speed (WS). The experiment was carried out for a total of 12 soil samples of two soil types (clay/sand) and two compaction levels (compacted/non-compacted). After data analysis, calibration models relating soil moisture content (SMC) to differential temperature (Td), relative humidity (RH), solar radiation (SR), and wind speed (WS) were generated using stepwise multiple linear regression of the calibration data set. The performance of the models was evaluated using validation data. Four mathematical models of predicting soil moisture content were generated for each soil type and configuration using the calibration data set. Among the four models, the best model for each soil type and configuration was determined by comparing root mean of squared errors of calibration (RMSEC) and root mean of squared errors of validation (RMSEV) values. Furthermore, a calibration model for the thermal infrared sensor was developed to determine the corrected soil surface temperature as measured by the sensor (Tir) instead of using the thermocouples. The performance of the thermal infrared sensor to predict soil moisture content was then tested for sand compacted and sand non-compacted soils and compared to the predictive performance of the thermocouples. This was achieved by using the measured soil surface temperature by the sensor (Tir), instead of the measured soil surface temperature using the thermocouples to determine the soil-minus-air temperature (Td). The sensor showed comparable prediction performance, relative to thermocouples. <p>Overall, the models developed in this study showed high prediction performance when tested with the validation data set. The best models to predict SMC for compacted clay soil, non-compacted clay soil, and compacted sandy soil were three-variable models containing three predictive variables; Td, RH, and SR. On the other hand, the best model to predict SMC for compacted sandy soil was a two-variable model containing Td, and RH. The results showed that the prediction performance of models for predicting SMC for the sandy soils was superior to those of clay soils.

An investigation of the high spectral resolution signatures and radiative forcing of tropospheric aerosol in the thermal infrared was conducted. To do so and to support advanced modeling of optical properties, a high spectral resolution library of atmospheric aerosol optical constants was developed. This library includes new optical constants of sulfate-nitrate-ammonium aqueous solutions and the collection of a broad range of existing optical constants for aerosol components, particularly mineral optical constants. The mineral optical constants were used to model and study infrared dust optical signatures as a function of composition, size, shape and mixing state. In particular, spherical and non-spherical optical models of dust particles were examined and compared to high spectral resolution laboratory extinction measurements. Then the performance of some of the most common effective medium approximations for internal mixtures was examined by modeling the optical constants of the newly determined sulfate-nitrate-ammonium mixtures. The optical signature analysis was applied to airborne and satellite high spectral resolution thermal infrared radiance data impacted by Saharan dust events. A new technique to retrieve dust microphysical properties from the dust spectral signature was developed and compared to a standard technique. The microphysics retrieved from this new technique and from a standard technique were then used to investigate the effects of dust on radiative forcing and cooling rates in the thermal IR.

La télédétection en imagerie hyperspectrale infrarouge thermique est l'étude d'images en luminance, acquises depuis un avion ou un satellite dans le domaine spectral de l'infrarouge thermique. Ces images sont liées à l'émissivité et à la température, estimées par les méthodes de découplage température/émissivité (T/E), ainsi qu'à l'abondance, estimée par les méthodes de démélange, des matériaux présents dans la scène. Si les méthodes de découplage T/E ont été largement étudiées, les méthodes de démélange dans ce domaine spectral restent peu explorées : c'est l'objectif de cette thèse. Pour cela, nous avons mis en place trois stratégies de démélange. Dans un premier temps, le démélange est effectué sur les luminances. Cette stratégie donne globalement de bons résultats mais est relativement sensible aux variations spatiales de la température. La deuxième stratégie, démélangeant à partir des estimations d'émissivité des méthodes de découplage T/E, s'affranchit de cette variation spatiale mais donne des résultats plus bruités. Enfin, une méthode de démélange basée sur l'estimation conjointe de la température et des abondances a été élaborée. Cette méthode s'appelle Thermal Remote sensing Unmixing for Subpixel Temperature (TRUST) et donne de meilleurs résultats que la première stratégie tout en étant robuste aux variations spatiales de la température. / Thermal hyperspectral remote sensing provides information about materials from the measured radiance image. It is achieved using temperature and emissivity separation (TES) methods, estimating the emissivity and the temperature of the materials, and using unmixing methods, estimating their abundances. TES methods have been well investigated while too few studies have been working on unmixing in thermal infrared domain : this is the objective of this PhD. Therefore, three strategies have been studied. First, the unmixing is applied on radiance. It achieves good results but depends on the spatial variation of temperature. Applying the unmixing on the emissivities, estimated using the TES methods, gets rid of the spatial variation of temperature but provides a noisy abundance estimation. Eventually, a new method called Thermal Remote sensing Unmixing for Subpixel Temperature (TRUST) is designed to jointly estimate the abundance and the temperature of materials within the pixels. It gives better results than the first strategy and is more robust to spatial variation of temperature.

Now and then train accidents occur. Collisions between trains and objects such as animals, humans, cars, and fallen trees can result in casualties, severe damage on the train, and delays in the train traffic. Thus, train collisions are a considerable problem with consequences affecting society substantially. The company Termisk Systemteknik AB has on commission by Rindi Solutions AB investigated the possibility to detect anomalies on the railway using a trainmounted thermal imaging camera. Rails are also detected in order to determine if an anomaly is on the rail or not. However, the rail detection method does not work satisfactory at long range. The purpose of this master’s thesis is to improve the previous rail detector at long range by using machine learning, and in particular deep learning and a convolutional neural network. Of interest is also to investigate if there are any advantages using cross-modal transfer learning. A labelled dataset for training and testing was produced manually. Also, a loss function tailored to the particular problem at hand was constructed. The loss function was used both for improving the system during training and evaluate the system’s performance during testing. Finally, eight different approaches were evaluated, each one resulting in a different rail detector. Several of the rail detectors, and in particular all the rail detectors using crossmodal transfer learning, perform better than the previous rail detector. Thus, the new rail detectors show great potential to the rail detection problem.

Automatic tracking of an object of interest in a video sequence is a task that has been much researched. Difficulties include varying scale of the object, rotation and object appearance changing over time, thus leading to tracking failures. Different tracking methods, such as short-term tracking often fail if the object steps out of the camera’s field of view, or changes shape rapidly. Also, small inaccuracies in the tracking method can accumulate over time, which can lead to tracking drift. Long-term tracking is also problematic, partly due to updating and degradation of the object model, leading to incorrectly classified and tracked objects. This master’s thesis implements a long-term tracking framework called Tracking-Learning-Detection which can learn and adapt, using so called P/N-learning, to changing object appearance over time, thus making it more robust to tracking failures. The framework consists of three parts; a tracking module which follows the object from frame to frame, a learning module that learns new appearances of the object, and a detection module which can detect learned appearances of the object and correct the tracking module if necessary. This tracking framework is evaluated on thermal infrared videos and the results are compared to the results obtained from videos captured within the visible spectrum. Several important differences between visual and thermal infrared tracking are presented, and the effect these have on the tracking performance is evaluated. In conclusion, the results are analyzed to evaluate which differences matter the most and how they affect tracking, and a number of different ways to improve the tracking are proposed.

Les données de télédétection dans l'infrarouge thermique (IRT) sont une source indispensable d'information pour estimer les flux de surface et suivre le fonctionnement des agro-écosystèmes. Cependant, les mesures de température de surface sont sujettes à des effets directionnels très importants (présence de 'hot spot') pouvant entraîner une erreur allant jusqu'à une dizaine de degrés Celsius. Ils doivent être pris en compte en vue des applications opérationnelles. Le travail proposé ici vise à modéliser l'anisotropie directionnelle des couverts végétaux pour mettre au point des méthodes opérationnelles de correction des mesures satellitaires de température de surface. Il est largement motivé par les projets du CNES visant à élaborer une mission spatiale nouvelle combinant une haute résolution spatiale et des capacités fortes de revisite dans l'IRT. Deux étapes de travail ont été menées. La première repose sur l'utilisation du modèle déterministe de transfert Sol-Végétation-Atmosphère SCOPE (Soil Canopy Observation, Photochemistry and Energy fluxes), capable de simuler les radiances directionnelles dans l'optique et l'IRT. Dans ce manuscrit, il est validé par rapport à des mesures de terrain et sa capacité à simuler correctement les effets d'anisotropie démontrée. Il est ensuite utilisé pour étudier de façon systématique la sensibilité de l'anisotropie directionnelle à la structure de la canopée, à son état hydrique, au forçage météorologique et aux configurations angulaires solaire et de visée. Les conséquences en terme d'impact combiné des caractéristiques orbitales des satellites, de la position géographique des sites observés et de la date d'acquisition sur l'anisotropie sont discutées. La seconde étape vise à proposer un modèle paramétrique simplifié (dit RL). SCOPE est ici utilisé en tant que générateur de données. Le modèle RL se révèle robuste et capable de restituer avec succès les signatures directionnelles sur le plan géométrique (position du hot spot) comme pour l'amplitude des effets directionnels. Une comparaison avec le seul autre modèle paramétrique utilisé jusqu'alors en télédétection IRT (le modèle de Vinnikov) confirme les qualités du modèle RL, ce qui en fait un candidat potentiel pour les chaines de traitement des futures données satellitaires. / Remotely-sensed data in thermal infrared (TIR) are an essential source of information to estimate surface fluxes and to monitor the functioning of agro-ecosystems. However, surface temperature measurements are prone to directional effects ('hot spot' phenomenon)which may result in an error up to 10°C. They have to be taken into account in the framework of operational applications. The work proposed here aims at modelling the directional anisotropy of continuous vegetated canopies in order to develop operational methods for correcting land surface temperature measurements carried out by TIR satellites. This work is mainly motivated by the CNES projects aiming at developing a new TIR spatial mission combining both high spatial resolution and high revisit time capacities. Two steps were carried out. The first is based on the use of the deterministic SVAT model SCOPE (Soil Canopy Observation, Photochemistry and Energy fluxes), able to simulate directional radiances at top of canopy in both optical and TIR domains. In this thesis, it is validated against experimental measurements and its ability to successfully simulate TIR directional anisotropy demonstrated. Then it is used to study the sensitivity of anisotropy to canopy structure, water status of soil and vegetation, meteorological forcing and solar and observer angular configurations. The consequences of the combined features of satellites orbits, geographical position of the scanned sites and acquisition date on anisotropy are discussed. In the second part, we propose a simplified parametric model (called 'RL'). SCOPE is used as a data generator. The RL model is deemed suitable and able to correctly reproduce directional signatures both in terms of geometry (hot spot position) and amplitude of these effects. A comparison with the only one parametric model previously used in TIR remote sensing (Vinnikov's approach) confirms the good capacities of the RL model. The RL model is thus a potential candidate to the future satellite processing chains.