The aim of this study was to detect spatio-temporal changes in sugarcane land use using satellite imagery for 1991–2006 in Umbumbulu, South Africa. This change detection study will enable quantification of change and the changes between different land use and land cover that has occurred over the study period 1991–2006. This work embarked on a change detection analysis using image-processing software namely ERDAS, IDRISI and ArcGIS to complete the study. Three Landsat TM images from 1991, 2001, and 2006 were used. The images were geometrically corrected to a common map projection, followed by image processing operations namely: radiometric correction, supervised image classification, accuracy assessment and post classification comparison change detection. Each image was separately classified into land cover categories of water, grassland, mix bush/shrub, forestry, sugarcane and built-up land using the supervised classification maximum likelihood algorithm in ERDAS. Final classification accuracy was determined to be ‘satisfactory’ or ‘good’ by means of employing standardized accuracy assessment measures, the error matrix. The post-classification comparison technique was applied to compare the classified images to assess for changes in sugarcane land use over time using IDRISI software. The classified images produced were exported into ArcMap GIS software for additional change analysis. The results are displayed as change maps. Change analysis has been executed based on digital interpretation of classification results. / Thesis (M.Env.Dev.) - University of KwaZulu-Natal, Pietermaritzburg, 2009.

Leaf area index (LAI) is an important variable in models that attempt to simulate carbon, nutrient, water and energy fluxes for forest ecosystems. LAI can be measured either directly (destructive sampling) or by using indirect techniques that involve estimation of LAI from light penetration through canopies. Destructive sampling techniques are laborious, expensive and can only be carried out for small plots. Although indirect techniques are non-destructive and less time consuming, they assume a random foliage distribution that rarely occurs in nature. Thus a technique is required that would allow for rapid estimation of LAI at the stand level. A means of getting this information is via remotely sensed measurements of reflected energy with an airborne or satellite-based sensor. Such information on an important plant species such as Acacia mearnsii (Black Wattle) is vital as it provides an insight into its water use. Landsat ETM+ images covering four study sites In KwaZulu-Natal midlands encompassing pure stands of Acacia mearnsii were processed to obtain four types of vegetation indices (VIs). The indices included: normalized difference vegetation index (NDVI), ratio vegetation index (RVI), transformed vegetation index (TVI) and vegetation index 3 (VB). Ground based measurements of LAI were made using destructive sampling (actual LAI) and LAI-2000 optical instrument, (plant area index, PAl). Specific leafarea (SLA) and leaf area (LA) were measured in the field for the entire sample stands to estimate their LAI values. The relationships between the various VIs and SLA, actual LAI and PAl values measured by LAI-2000 were evaluated using correlation and regression statistical analyses. Results showed that the overall mean SLA value of Acacia mearnsii was 8.28 m2kg-1 SLA showed strong correlations with NDVI (r=0.71, p<O.Ol) and RVI (r=0.76, p<O.Ol) and a moderate correlation with TVI (r=0.66, p<0.05). Regression analysis revealed that SLA had significant relationship with RVI (R2=0.59) and NDVI (R2=0.51). Actual LAI values showed strong correlation with PAl values (r=0.86) and the analysis revealed that 74 % of the variation in the relationship between actual LAI and PAl values could be explained by regression. PAl values were strongly correlated with NDVI (r=0.75,p<O.Ol) and moderately correlated with RVI (r=O.63, p<O.05) and TVI (r=O.58, p<O.05). Actual LAI was strongly correlated with NDVI (r=O.79, p<O.Ol) and moderately correlated with RVI (r=O.61, p<O.05). Out of the various VIs examined in this study, NDVI was found to have a better relationship with actual LAI values (R2=O.62) and with PAI values (R2=O.56); while VB didn't show any significant relationship with SLA, PAl or actual LAl. In conclusion, preliminary estimate of SLA of Acacia mearnsii could be obtained from RVI or NDVl. The relationship obtained between PAl and actual LAI values was satisfactory, thus the regression equation can be used to calibrate the LAI-2000 plant canopy analyzer. Because NDVI was observed to have a good relationship with actual LAI and PAl, LAI of Acacia mearnsii can be estimated from Landsat ETM+ satellite imagery with a reasonable degree of accuracy. These results can satisfactorily be used as inputs into models that attempt to estimate water use by Acacia mearnsii. / Thesis (M.Sc.)-University of Natal, Pietermaritzburg, 2003.

Remote sensing offers major contributions to the understanding of northern landscapes and vegetation patterns. Recently available instrumentation and analytic techniques, yielding new types of data and new approaches to longstanding problems, are demonstrated in this analysis of terrain conditions and vegetation distributions in the Caribou Hills, N.W.T. The analysis of landform was based on field data, image interpretation and photogrammetric elevation model data. Slope angles and aspects were computed and trend surfaces, residuals and contour maps produced for model areas. Within sampled areas, surface roughness, the degree of dissection and the apparent dominance of either fluvial or mass wasting processes were found to be controlled by slope aspect, snow drifting patterns and the nival melt schedule. Patterns of active layer depth and details of surface materials, morphology and processes were derived from stereoscopic analysis of photographs through linkages with plant associations. Twelve plant associations, defined by field survey, provided a basis for differentiating photographic signatures and vegetation mapping classes. The character and separability of the spectral signatures were reviewed using ratioed and clustered optical film density data. The major advantages of remote sensing as an analytic tool were demonstrated. Remote sensing provides a vast array of geographic data and a unique synthesis of terrain and vegetation conditions offering the researcher key information that is otherwise unavailable.

The aim of this thesis is the modeling of surface directional thermal radiation and angular correction on the LST by using empirical and physical methods as well as the analysis of field validation. The work has conducted to some conclusions. The directional emissivity of natural surfaces was obtained from MODIS emissivity product and then used in the split-window algorithm for angular correction on LST. The parameterization models of directional emissivity and thermal radiation were developed. As for the non-isothermal pixels, the daytime-TISI method was proposed to retrieve directional emissivity and effective temperature from multi-angular middle and thermal infrared data. This was validated using an airborne dataset. The kernel-driven BRDF model was checked in the thermal infrared domain and its extension was used to make angular normalization on the LST. A new model, namely FovMod that concerns on the footprint of ground sensor, was developed to simulate directional brightness temperature of row crop canopy. Based on simulation result of the FovMod, an optimal footprintfor field validation of LST was obtained. This thesis has systematically investigated the topic of directional thermal radiation and angular correction on surface temperature and its findings will improve the retrieval accuracy of temperature and emissivity from remotely sensed data and will also provide suggestion for the future design of airborne or spaceborne multi-angular thermal infrared sensors and also for the ground measurement of surface parameters.

It is becoming increasingly important to monitor biodiversity in the extensive Australian rangelands; currently however, there is no method capable of achieving this goal. There are two potential sources of relevant data that cover the Australian rangelands, and from which measures of biodiversity might be extracted: traditional field-based methods such as quadrat surveys have collected flora and fauna species data throughout the rangelands, but at fine scale; satellite remote sensing collects biologically relevant, spatially comprehensive data. The goal of this thesis was to provide the spatially comprehensive measure of biodiversity required for informed management of the Australian rangelands. The study specifically focused on the Stony Plains in the South Australian rangelands. To that end the thesis aimed to develop indices capable of measuring and/or monitoring biodiversity from vegetation quadrat survey data and remotely sensed data. The term biodiversity is so all-encompassing that direct measurement is not possible; therefore it is necessary to measure surrogates instead. Total perennial vegetation species richness (y-diversity) is a sound surrogate of biodiversity: the category of species is well defined, species richness is measurable, and there is evidence that vegetation species richness co-varies with the species richness of other taxonomic groups in relation to the same environmental variables. At least two broad scale conventional vegetation surveys are conducted in the study region; the Biological Survey of South Australia; and the South Australian Pastoral Lease Assessment. Prior to the extraction of biodiversity data the quality of the BSSA, the best biodiversity survey, was evaluated. Analysis revealed that false-negative errors were common, and that even highly detectable vegetation species had detection probabilities significantly less than one. Without some form of correction for detectability, the species diversity recorded by either vegetation survey must be treated with caution. Informed by the identification of false-negative errors, a method was developed to extract y-diversity of woody perennials from the survey data, and to remove the influence of sampling effort. Data were aggregated by biogeographic region, rarefaction was used to remove most of the influence of sampling effort, and additional correction removed the residual influence of sampling effort. Finally, additive partitioning of species diversity allowed extraction of indices of a-, β- and y-diversity free from the influence of sampling effort. However, this woody perennial vegetation y-diversity did not address the need for a spatially extensive, fine scale measure of biodiversity at the extent of the study region. The aggregation of point data to large regions, a necessary part of this index, produces spatially coarse results. To formulate and test remotely sensed surrogates of biodiversity, it is necessary to understand the determinants of and pressures on biodiversity in the Australian rangelands. The most compelling explanation for the distribution of biodiversity at the extensive scales of the Australian rangelands is the Productivity Theory, which reasons that the greater the amount and duration of primary productivity the greater the capacity to generate and support high biodiversity. The most significant pressure on biodiversity in the study area is grazing-induced degradation, or overgrazing. Two potential spatially comprehensive surrogates of pressure on biodiversity were identified. The first surrogate was based on the differential effect of overgrazing on waterenergy balance and net primary productivity: water-energy balance is a function of climatic variables, and therefore a measure of potential or expected primary productivity; net primary productivity is reduced by high grazing pressure. The second surrogate was based on the effect of grazing-induced degradation on the temporal variability of net primary productivity: overgrazing reduces mean net primary productivity and rainfall use efficiency, and increases variation in net primary productivity and rainfall use efficiency. The two surrogates of biodiversity stress were derived from the best available remotely sensed and climate data for the study area: actual evapotranspiration recorded by climate stations was considered an index of water-energy balance; net primary productivity was measured from NOAA AVHRR integrated NDVI; rainfall use efficiency (biomass per unit rainfall) was calculated from rainfall data collected at climate stations and the net primary productivity measure. Finally, the surrogates were evaluated against the index of woody perennial a-, β- and y-diversity, on the assumption that prolonged biodiversity stress would reduce vegetation species diversity. No link was found between Surrogate 1 and woody perennial a-, β- or y-diversity. The relationship of Surrogate 2 to woody perennial diversity was more complex. Only some of the results supported the hypothesis that overgrazing decreases y-diversity and average NPP and RUE. Importantly, none of the results supported the most important part of the hypothesis that the proposed indices of biodiversity pressure would co-vary with woody perennial a-diversity. Thus, the analysis did not reveal a convincing link between either surrogate and vegetation species diversity. However, the analysis was hampered to a large degree by the climate data, which is interpolated from a very sparse network of climate stations. This thesis has contributed significantly to the measurement and monitoring of biodiversity in the Australian rangelands. The identification of false-negative errors as a cause for concern will allow future analyses of the vegetation survey data to adopt methods to counteract these errors, and hence extract more robust information. The method for extracting sampling effort corrected indices of a-, β- and y-diversity allow for the examination and comparison of species diversity across regions, regardless of differences in sampling effort. These indices are not limited to rangelands, and can be extracted from any vegetation quadrat survey data obtained within a prescribed methodology. Therefore, these tools contribute to global biodiversity measurement and monitoring. Finally, the remotely sensed surrogates of biodiversity are theoretically sound and applicable in any rangeland where over-grazing is a significant source of degradation. However, because the evaluation of these surrogates in this thesis was hampered by available data, further testing is necessary. / http://proxy.library.adelaide.edu.au/login?url= http://library.adelaide.edu.au/cgi-bin/Pwebrecon.cgi?BBID=1346544 / Thesis (Ph.D.) - University of Adelaide, School of Earth and Environmental Sciences, 2008

Thesis (MSc (Geography and Environmental Studies))--University of Stellenbosch, 2005. / South Africa’s natural vegetation and soils, which are essential resources for agricultural practices, are becoming degraded. Natural resource disturbances can also cause extensive harm to local communities and their economies. To allow successful natural resource monitoring, there is an urgent need for integrated GIS spatial data and development of remotely sensed indicators of key ecosystems processes. Satellite remote sensing provides the most cost-effective and reliable tool for generating these spatial data. The main objective of the study is, therefore, to develop and evaluate methodologies for assessing, mapping and monitoring the condition of natural resources in southern Africa with the aid of remote sensing and GIS. The resulting integrated spatial framework represents methodologies for, firstly, identifying and accessing vegetation and soil parameters on a gradient from pristine to degraded condition; secondly, identifying, assessing, processing and modelling GIS and remotesensing spatial data to derived degradation maps, which identify rangeland condition and woody cover classes and, thirdly, comparing two satellite remote-sensing sensors (LANDSAT ETM and MODIS) and making statements of degradation. This approach could make an integrated spatial framework comprehensive in its considerations of provincial degradation mapping and robust enough to be used for monitoring on a national scale. By acquiring spatial and non-spatial data in a quantitative logically robust but accurate manner, integrated spatial frameworks provides the structure for combining specialized information as well as for analysis in an effective management programme. This could guide rangeland managers in assessing, mapping and monitoring of natural resources in a scientifically acceptable way. All of these factors emphasise the need for the development of a national rangeland monitoring strategy and monitoring system.

Thesis (MA.)--Stellenbosch University, 2001. / ENGLISH ABSTRACT: There are many reasons to conserve fynbos. Not only does fynbos form part of the Cape floral kingdom, one of the richest floral kingdoms in the world, but the contribution that it makes to the regional economy through utilisation, education, recreation and tourist opportunities is immeasurable. Fire plays an integral role in fynbos ecosystems. According to Van Wilgen, Richardson & Seydack (1994: 322) " ... managing fynbos equates to managing fire". Therefore managers need accurate fire information about a fynbos area to manage it properly. This is where satellite remote sensing can provide the manager with useful information about the fire regime. In other words, satellite remote sensing can help a manager establish where and when an area has burnt. Using readily available satellite data, this study attempts to establish (through comparison) what techniques would be most suitable and affordable to compile a fire information database. Landsat Thematic Mapper data from 1990 - 1996 of the southwestern Cape was used and compared with existing fire records of the area. The results show that techniques such as supervised and unsupervised classification are reliable in identifying burnt areas, but a major drawback of these techniques is that they require a large amount of user input and knowledge. They are thus not regarded as simple or easily repeatable. - The' more simple techniques like image differencing and image ratioing were also found to be reliable in identifying burnt areas. These techniques require less user input and in some instances less data (image bands) to produce similar (or better) results than supervised and unsupervised classification techniques. The results show that differencing temporally different Images, obtained from applying principle components analysis, produces reliable results with very little confusion and little user input. Using such a technique could enable users to procure only two bands of Landsat data and still produce reliable fire information for managing a fynbos ecosystem. / AFRIKAANSE OPSOMMING: Daar is verskeie redes waarom fynbos bewaar moet word. Nie net vorm dit deel van een van die rykste blommeryke in die wereld nie, maar die bydrae wat dit tot die streeksekonomie maak, deur die benutting van veldblomme en die geleenthede wat dit bied vir toerisme en ontspanning, is enorm. Vuur speel 'n belangrike rol in die bestuur van fynbos ekosisteme. Soos beklemtoon deur Van Wilgen, Richardson & Seydack (1994: 322) se stelling: " ... managing fynbos equates to managing fire". Om hierdie rede is dit belangrik dat 'n bestuurder akkurate inligting oor die verspreiding van veldbrande moet he. Satellietafstandwaarneming kan hier 'n belangrike rol speel deur sulke inligting te verskaf Deur gebruik te maak van maklik bekombare satellietdata, poog hierdie studie om te bepaal (d.m.v. vergelyking) watter tegnieke die mees geskikte is in terme van bekostigbaarheid en gebruikersvriendelikheid. Landsat Thematic Mapper data van 1990 tot 1996 van die suidwes-Kaap is gebruik en vergelyk met bestaande branddata van die studiegebied. Daar is gevind dat tegnieke soos gerigte en nie-gerigte klassifikasie in staat is om gebrande dele betroubaar uit te ken. Hierdie tegnieke verg egter baie insette en kennis van die gebruiker en is ook nie maklik om jaar na jaar te herhaal nie. Daarom word hierdie tegnieke nie aanbeveel nie. Daar is gevind dat die eenvoudiger tegnieke soos veranderingsanalise ook gebrande dele betroubaar kon uitken. Hierdie tegnieke het die voordeel dat die gebruiker nie baie' kennis van die gebied hoef te he nie en ook nie so baie insette hoef te lewer nie. Hierdie tegnieke word bo gerigte en nie-gerigte klassifikasie aanbeveel. - Die resultate dui daarop dat betroubare resultate verkry kan word deur tempo reel verskillende beeIde, verkry deur hoofkomponentanalise, van mekaar af te trek. Hierdie tegniek vereis relatief min gebruikersinsette en daar kan selfs met slegs twee Landsat bande gewerk word. So 'n tegniek kan beslis 'n bekostigbare en effektiewe manier wees om nodige inligting vir die bestuur van 'n fynbos ekosisteem te bekom.

Made available in DSpace on 2016-12-23T13:51:49Z (GMT). No. of bitstreams: 1 Dissertacao de Telma Machado de Oliveira Peluzio.pdf: 3946411 bytes, checksum: c58d6b9d249683aaab8229a9b18fc473 (MD5) Previous issue date: 2010-09-17 / Aracruz Celulose S.A. / Com esta pesquisa avaliou-se o resultado de diferentes metodologias na determinação do uso e ocupação da terra e áreas de preservação permanente utilizando aerofotos digitais, com resolução espacial de 1m (cenário 01) e imagens de satélite, com resolução espacial de 0,5m (cenário 02), fotointerpretadas em tela, na escala cartográfica de 1:2.000, da bacia hidrográfica do córrego Horizonte, Alegre-ES. Foram avaliadas as áreas de preservação permanente ao longo dos cursos d água, entorno de nascentes, terrenos com declividade igual ou superior a 45 graus e terço superior de topo de morros para os cenários 01 e 02. As áreas que deveriam ser destinadas as áreas de preservação permanentes no cenário 01 e 02 totalizam 5,97km² e 5,87km², respectivamente, diferindo apenas nas APPs em torno de nascentes (0,13km² e 0,10km²) e, APPs ao longo dos cursos d água (1,87km² e 1,79km²), não havendo diferença para as APPs de declividade igual ou superior a 45 graus (0,04 km²) e APPs de terço superior de topo de morro (3,94km²). Utilizouse os índices de desempenho global e índice Kappa para determinação do erro médio entre os cenários e teste t a 5% de probabilidade para avaliar o grau de significância no processo de fotointerpretação entre os cenários 01 e 02. Das 27 classes de uso da terra identificadas, a pastagem e fragmento florestal ocupam 45,54% e 24,26% no cenário 01 e, 33,50% e 19,54% no cenário 02, respectivamente. Do total de áreas de preservação permanentes determinadas no cenário 01, apenas 33,92% das APPs ocupam essa função, enquanto no cenário 02, esse percentual é de 35,92%. A diferença na fotointerpretação dos cenários em estudo deve-se às bandas vermelho, verde e azul (cenário 01) e vermelho, verde, azul e infravermelho próximo (cenário 02). Verifica-se uma variação positiva do desempenho global em 6,67% do cenário 01 em relação ao cenário 02, e variação positiva do índice Kappa em 2,09%, do cenário 02 em relação ao cenário 01, não apresentando significância pelo teste t no processo de fotointerpretação entre a aerofoto e a imagem de satélite / With this study we evaluated the result of different methodologies in determining the use and occupancy of land and permanent preservation areas using digital aerial photographs with spatial resolution of 1m (scenario 01) and satellite imagery with a spatial resolution of 0.5 m ( scenario 02), photointerpreted screen, on a scale of 1:2000 mapping, watershed stream Horizonte, Alegre-ES. We avalueted the permanent preservation araes along watercourses, around springs, land with slopes greater than 45 degrees and upper third of the top of hills for scenarios 01 and 02. Areas that should be destined to permanent preservation areas in the scene 01 and 02 totaling 5.97 km ² and 5.87 km ², respectively, differing only in APPs around springs (0.13 km ² 0.10 km ²) and over the APPs watercourses (1.87 km ² 1.79 km ²), with no difference for APPs slope less than 45 degrees (0.04 km ²) APP and the upper third of the hilltop (3.94 km ²). We used the global performance indices and Kappa index for determining the average error between scenarios 01 and 02. Of the 27 classes of land use identified, grassland and forest fragmentation occupy 45.54% and 24.26% in stage 01, and 33.50% and 19.54% in stage 02, respectively. Of the total of permanent preservation areas in certain stage 01, only 33.92% of APPs occupy that role, while in scenario 02, that percentage is 35.92%. The difference in photo-interpretation of the scenarios under study is due to the bands red, green and blue (scenario 01) and red, green, blue and near infrared (scenario 02). There is a positive change in the overall performance of the scenario 01 6.67% compared to 2002 scenario, and positive change in the kappa index 2.09%, the scenario in 2002 compared to 2001 scenario, showing no significance by statistical analysis at 5% level by t test.

Algal blooms plague freshwaters across the globe, as increased nutrient loads lead to eutrophication of inland waters and the presence of potentially harmful cyanobacteria. In this context, remote sensing is a valuable approach to monitor water quality over broad temporal and spatial scales. However, there remain several challenges to the accurate retrieval of water quality parameters, and the research in this thesis investigates these in an optically complex lake (Lake Balaton, Hungary). This study found that bulk and specific inherent optical properties [(S)IOPs] showed significant spatial variability over the trophic gradient in Lake Balaton. The relationships between (S)IOPs and biogeochemical parameters differed from those reported in ocean and coastal waters due to the high proportion of particulate inorganic matter (PIM). Furthermore, wind-driven resuspension of mineral sediments attributed a high proportion of total attenuation to particulate scattering and increased the mean refractive index (n̅p) of the particle assemblage. Phytoplankton pigment concentrations [chlorophyll-a (Chl-a) and phycocyanin (PC)] were also accurately retrieved from a times series of satellite data over Lake Balaton using semi-analytical algorithms. Conincident (S)IOP data allowed for investigation of the errors within these algorithms, indicating overestimation of phytoplankton absorption [aph(665)] and underestimation of the Chl-a specific absorption coefficient [a*ph(665)]. Finally, Chl-a concentrations were accurately retrieved in a multiscale remote sensing study using the Normalized Difference Chlorophyll Index (NDCI), indicating hyperspectral data is not necessary to retrieve accurate pigment concentrations but does capture the subtle heterogeneity of phytoplankton spatial distribution. The results of this thesis provide a positive outlook for the future of inland water remote sensing, particularly in light of contemporary satellite instruments with continued or improved radiometric, spectral, spatial and temporal coverage. Furthermore, the value of coincident (S)IOP data is highlighted and contributes towards the improvement of remote sensing pigment retrieval in optically complex waters.

Orientadores: Ricardo da Silva Torres, Alexandre Xavier Falcão / Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-23T15:18:27Z (GMT). No. of bitstreams: 1 Santos_JeferssonAlexdos_D.pdf: 18672412 bytes, checksum: 58ac60d8b5342ab705a78d5c82265ab8 (MD5) Previous issue date: 2013 / Resumo: Um grande esforço tem sido feito para desenvolver sistemas de classificação de imagens capazes de criar mapas temáticos de alta qualidade e estabelecer inventários precisos sobre o uso do solo. As peculiaridades das imagens de sensoriamento remoto (ISR), combinados com os desafios tradicionais de classificação de imagens, tornam a classificação de ISRs uma tarefa difícil. Grande parte dos desafios de pesquisa estão relacionados à escala de representação dos dados e, ao mesmo tempo, à dimensão e à representatividade do conjunto de treinamento utilizado. O principal foco desse trabalho está nos problemas relacionados à representação dos dados e à extração de características. O objetivo é desenvolver soluções efetivas para classificação interativa de imagens de sensoriamento remoto. Esse objetivo foi alcançado a partir do desenvolvimento de quatro linhas de pesquisa. A primeira linha de pesquisa está relacionada ao fato de embora descritores de imagens propostos na literatura obterem bons resultados em várias aplicações, muitos deles nunca foram usados para classificação de imagens de sensoriamento remoto. Nessa tese, foram testados doze descritores que codificam propriedades espectrais e sete descritores de textura. Também foi proposta uma metodologia baseada no classificador K-Vizinhos mais Próximos (K-nearest neighbors - KNN) para avaliação de descritores no contexto de classificação. Os descritores Joint Auto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition (SID) e Quantized Compound Change Histogram (QCCH), apresentaram os melhores resultados experimentais na identificação de alvos de café e pastagem. A segunda linha de pesquisa se refere ao problema de seleção de escalas de segmentação para classificação de imagens de sensoriamento baseada em objetos. Métodos propostos recentemente exploram características extraídas de objetos segmentados para melhorar a classificação de imagens de alta resolução. Entretanto, definir uma escala de segmentação adequada é uma tarefa desafiadora. Nessa tese, foram propostas duas abordagens de classificação multiescala baseadas no algoritmo Adaboost. A primeira abordagem, Multiscale Classifier (MSC), constrói um classificador forte que combina características extraídas de múltiplas escalas de segmentação. A outra, Hierarchical Multiscale Classifier (HMSC), explora a relação hierárquica das regiões segmentadas para melhorar a eficiência sem reduzir a qualidade da classificação xi quando comparada à abordagem MSC. Os experimentos realizados mostram que é melhor usar múltiplas escalas do que utilizar apenas uma escala de segmentação. A correlação entre os descritores e as escalas de segmentação também é analisada e discutida. A terceira linha de pesquisa trata da seleção de amostras de treinamento e do refinamento dos resultados da classificação utilizando segmentação multiescala. Para isso, foi proposto um método interativo para classificação multiescala de imagens de sensoriamento remoto. Esse método utiliza uma estratégia baseada em aprendizado ativo que permite o refinamento dos resultados de classificação pelo usuário ao longo de interações. Os resultados experimentais mostraram que a combinação de escalas produzem melhores resultados do que a utilização de escalas isoladas em um processo de realimentação de relevância. Além disso, o método interativo obtém bons resultados com poucas interações. O método proposto necessita apenas de uma pequena porção do conjunto de treinamento para construir classificadores tão fortes quanto os gerados por um método supervisionado utilizando todo o conjunto de treinamento disponível. A quarta linha de pesquisa se refere à extração de características de uma hierarquia de regiões para classificação multiescala. Assim, foi proposta uma abordagem que explora as relações existentes entre as regiões da hierarquia. Essa abordagem, chamada BoW-Propagation, utiliza o modelo bag-of-visual-word para propagar características ao longo de múltiplas escalas. Essa ideia foi estendida para propagar descritores globais baseados em histogramas, a abordagem H-Propagation. As abordagens propostas aceleram o processo de extração e obtém bons resultados quando comparadas a descritores globais / Abstract: A huge effort has been made in the development of image classification systems with the objective of creating high-quality thematic maps and to establish precise inventories about land cover use. The peculiarities of Remote Sensing Images (RSIs) combined with the traditional image classification challenges make RSI classification a hard task. Many of the problems are related to the representation scale of the data, and to both the size and the representativeness of used training set. In this work, we addressed four research issues in order to develop effective solutions for interactive classification of remote sensing images. The first research issue concerns the fact that image descriptors proposed in the literature achieve good results in various applications, but many of them have never been used in remote sensing classification tasks. We have tested twelve descriptors that encode spectral/color properties and seven texture descriptors. We have also proposed a methodology based on the K-Nearest Neighbor (KNN) classifier for evaluation of descriptors in classification context. Experiments demonstrate that Joint Auto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition (SID), and Quantized Compound Change Histogram (QCCH) yield the best results in coffee and pasture recognition tasks. The second research issue refers to the problem of selecting the scale of segmentation for object-based remote sensing classification. Recently proposed methods exploit features extracted from segmented objects to improve high-resolution image classification. However, the definition of the scale of segmentation is a challenging task. We have proposed two multiscale classification approaches based on boosting of weak classifiers. The first approach, Multiscale Classifier (MSC), builds a strong classifier that combines features extracted from multiple scales of segmentation. The other, Hierarchical Multiscale Classifier (HMSC), exploits the hierarchical topology of segmented regions to improve training efficiency without accuracy loss when compared to the MSC. Experiments show that it is better to use multiple scales than use only one segmentation scale result. We have also analyzed and discussed about the correlation among the used descriptors and the scales of segmentation. The third research issue concerns the selection of training examples and the refinement of classification results through multiscale segmentation. We have proposed an approach for xix interactive multiscale classification of remote sensing images. It is an active learning strategy that allows the classification result refinement by the user along iterations. Experimental results show that the combination of scales produces better results than isolated scales in a relevance feedback process. Furthermore, the interactive method achieves good results with few user interactions. The proposed method needs only a small portion of the training set to build classifiers that are as strong as the ones generated by a supervised method that uses the whole available training set. The fourth research issue refers to the problem of extracting features of a hierarchy of regions for multiscale classification. We have proposed a strategy that exploits the existing relationships among regions in a hierarchy. This approach, called BoW-Propagation, exploits the bag-of-visual-word model to propagate features along multiple scales. We also extend this idea to propagate histogram-based global descriptors, the H-Propagation method. The proposed methods speed up the feature extraction process and yield good results when compared with global low-level extraction approaches / Doutorado / Ciência da Computação / Doutor em Ciência da Computação