[Show description][Hide description] DESCRIPTION: The objective of this study was to evaluate if a multi-resolution segmentation algorithm is sensitive to the RapidEye’s Red Edge band and its benefits for vegetation mapping using GEOBIA and machine learning. We used a high-resolution multi-spectral RapidEye image taken in June, 2010. This image was segmented with a multiresolution segmentation algorithm (MRIS) using a fine scale parameter (300) and thirteen different weights (from ‘0’ to ‘100’) were assigned to the Red Edge spectral band to evaluate its influence in the segmentation and classification process. Each weight generated a segmented image. Attributes related to spectral information, geometry and texture were calculated for each image segment using the eCognition Developer®. Visual interpretation was performed along with field data to select seven classes (Dense vegetation, Meadow, Mining area, Bare land, Rock outcrop, Urban area and Water). A sample of 800 objects described by its attributes was selected from each segmented image. A decision tree approach based on CART was applied to the samples to select the attributes that provides the best separation among the classes within the scene. An accuracy assessment for the classification using CART was performed to compare the different weights assigned to the Red edge spectral band. Results showed that the Red Edge channel had no significant influence on the segmentation process. The attributes importance rank showed that the index derived from Red Edge channel can be used as input for image classification.
[Show abstract][Hide abstract] ABSTRACT: Remote sensing images are currently used as a source of auxiliary data for the inventory of native forests. These images when combined with geo-statistical techniques can provide gains in accuracy in inventory estimates. Accordingly, the aim of this study was to: (a) evaluate the spatial dependence structure of canopy reflectance values in a Cerrado Sensu Stricto fragment; (b) determine the correlations between the spectral and the wood volume data; (c) evaluate the pre-stratification efficiency based on the reflectance values from images of Landsat 5 TM satellite in a Cerrado fragment combined to kriging estimator and compare the random stratified sampling (RSS) estimates to systematic sampling (SS) estimates through the variable of interest in the forest inventory. The study area corresponds to a Cerrado Sensu Stricto fragment in Cônego Marinho city, MG. The forest inventory data were obtained from 41 plots distributed systematically. The wood volume was obtained by volumetric equations. The spectral data were collected from image in the satellite Landsat 5 TM. The spectral data were composed by TM1, TM2, TM3, TM4 and TM5 bands and the NDVI and SR vegetation indices. The spectral data have undergone a variographic analysis and were correlated with the total wood volume. Then, the stratification was carried out in the area from the spectral data. All the spectral variables showed spatially structured. The wood volume presented the highest correlations with the reflectance variables in TM4 band (r = -0,638) and reflectance in TM5 band (r = -0,501). The inventory error for SS was 19.11%, and ranged from 13.42% to 18.39% for different stratifications. Best stratifications were generated by spectral variables that presented the highest correlation values to the wood volume and those also presented the highest degree of spatial dependence (DE).
Scientia Forestalis/Forest Sciences 06/2015; 43(106):377-386. · 0.37 Impact Factor
[Show abstract][Hide abstract] ABSTRACT: The aim of this study was evaluate which spectral indices have higher separability in the discrimination of burned areas of other targets in the Landsat 5 TM images on Cerrado areas in northern Minas Gerais. For this, nine spectral indices were calculated; Burned Area Index (BAI), Char Soil Index (CSI), Enhanced Vegetation Index (EVI), Normalized Burn Ratio (NBR), variation of Normalized Burn Ratio (nbr2), Normalized Difference Vegetation Index (NDVI), Normalized Difference Moisture Index (NDMI), Mid-Infrared Burn index (MIRBI), Soil Adjusted Vegetation Index (SAVI) and checked your the separability by the index M. To assist the analysis, we calculated the basic statistics and the spectral signatures generated before and after the wildfire, and also the graph in space bi-spectral regions in red (band 3), near infrared (band 4), and shortwave infrared (bands 5 and 7). The results showed the MIRBI and NBR2 indices with better separability. Already NBR, NDVI, SAVI and CSI indices presented intermediate values of separability compared to other indices tested in this work. While the spectral indices BAI, EVI and NDMI had weak separability. We conclude that the indices using the bands of the shortwave infrared region showed the best results and are the most suitable for mapping fires in the region, Landsat 5 TM images.
[Show abstract][Hide abstract] ABSTRACT: The objective of this work was to determine the influence of stand age on the automatic detection of Eucalyptus sp. trees using LIDAR datasets. Three different stands 3, 5 and 7 years old were analyzed. The LIDAR cloud point data of the first return was split into two datasets: Class 1 (points for all vegetation), Class 2 (points for vegetation above 10m). Results for obtaining the number of stems for each dataset were compared to the census of the area, which was done by visual interpretation using an auxiliary high spatial resolution remote sensing image, and to forest inventory estimates. In comparison to the census data, tree counting using Class 1 dataset agreed well for all considered ages, with best results achieved in 3 and 5 year old stands. On the other hand, Class 2 biased toward underestimated values. The best results for this class were verified in 7 year old stands. When compared to the forest inventory data, this methodology proved to be more efficient. The number of stems derived from the forest inventory was biased towards overestimation. In order to achieve better estimates using forest inventory data, an intensification of the sampling procedure would be necessary.
[Show abstract][Hide abstract] ABSTRACT: O objetivo deste trabalho foi avaliar a possibilidade de se estimar o diâmetro à
altura do peito (DAP) com os dados de altura e de número de árvores derivados do
escâner a laser aerotransportado (LiDAR, "light detection and ranging"), e determinar
o volume de madeira de talhão de Eucalyptus sp. a partir dessas
variáveis. O número total de árvores detectadas foi obtido com uso da filtragem de
máxima local. A altura de plantas estimada pelo LiDAR apresentou tendência não
significativa à subestimativa. A estimativa do DAP foi coerente com os valores
encontrados no inventário florestal; porém, também mostrou tendência à subestimativa,
em razão do comportamento observado quanto à altura. A variável número de fustes
apresentou valores próximos aos observados nas parcelas do inventário. O LiDAR
subestimou o volume total de madeira do talhão em 11,4%, em comparação ao volume
posto na fábrica. A tendência de subestimação da altura das árvores (em média, cerca
de 5%) impactou a estimativa do volume individual de árvores e, consequentemente, a
do volume do talhão. No entanto, é possível gerar equações de regressão que estimam o
DAP com boa precisão, a partir de dados de altura de plantas obtidos pelo LiDAR. O
modelo parabólico é o que possibilita as melhores estimativas da produção volumétrica
dos talhões de eucalipto.
[Show abstract][Hide abstract] ABSTRACT: Este trabalho analisou a fragmentação florestal da Área de Proteção Ambiental Coqueiral, que está localizada no município de Coqueiral, região Sul do estado de Minas Gerais. O objetivo foi avaliar a fragmentação florestal da área de estudo, a partir de métricas da paisagem, bem como elaborar modelos de simulação da paisagem, no intuito de fornecer cenários futuros de restauração ecológica, e compará-los com a situação atual da paisagem. A análise do uso e ocupação da terra foi obtida por meio de técnicas de Sistemas de Informação Geográfica e Sensoriamento Remoto, a partir de uma imagem (SPOTMAP) do satélite SPOT 5. A análise da fragmentação florestal foi realizada utilizando o software FRAGSTATS, para calcular as métricas da paisagem mensurando parâmetros como: área, perímetro, forma, conectividade dos fragmentos. Para as simulações da paisagem foram criados buffers de 1 e 5 m no entorno de todos os remanescentes florestais da área de estudo, bem como a recuperação virtual das áreas de preservação permanente. A análise da fragmentação da paisagem mostrou que a vegetação natural está distribuída em 360 fragmentos, sendo 137 deles menores que 1 ha. Os modelos de simulação da paisagem mostraram que a área de vegetação aumentou de 1943,13 ha para 2299,02 ha na simulação em que as APPs foram reflorestadas (Vegetação natural/APPs restauradas = VA). O tamanho médio dos fragmentos nesta mesma simulação aumentou em relação à paisagem atual, passando de 7,66 m para 15,75 m. A paisagem VA mostrou um menor valor de forma (1,93), indicando que a forma dos fragmentos nesta simulação foi mais simples, o que é importante do ponto de vista da conservação, pois diminui o efeito de borda nos fragmentos. Os valores de isolamento não apresentaram diferença considerável nas simulações: 38,9 m (VN); 40,64 m (VB1); 42,89 m (VB5) e 39,75 m (VA), indicando um baixo isolamento dos fragmentos, mesmo na paisagem atual. O índice de conectividade foi alto (acima de 99%) para todas as simulações, indicando que as paisagens apresentam elevada conectividade estrutural. Estes dados são relevantes subsídios para a tomada de decisão e para gestão e planejamento da Área Proteção Ambiental Coqueiral, permitindo a indicação de áreas prioritárias para conservação.
[Show abstract][Hide abstract] ABSTRACT: The aim of this study was to develop a methodology for mapping land use and land cover in the northern region of Minas Gerais state, where, in addition to agricultural land, the landscape is dominated by native cerrado, deciduous forests, and extensive areas of vereda. Using forest inventory data, as well as RapidEye, Landsat TM and MODIS imagery, three specific objectives were defined: I) to test use of image segmentation techniques for an object-based classification encompassing spectral, spatial and temporal information, 2) to test use of high spatial resolution RapidEye imagery combined with Landsat TM time series imagery for capturing the effects of seasonality, and 3) to classify data using Artificial Neural Networks. Using MODIS time series and forest inventory data, time signatures were extracted from the dominant vegetation formations, enabling selection of the best periods of the year to be represented in the classification process. Objects created with the segmentation of RapidEye images, along with the Landsat TM time series images, were classified by ten different Multilayer Perceptron network architectures. Results showed that the methodology in question meets both the purposes of this study and the characteristics of the local plant life. With excellent accuracy values for native classes, the study showed the importance of a well-structured database for classification and the importance of suitable image segmentation to meet specific purposes.
[Show abstract][Hide abstract] ABSTRACT: Lands (broader concept than soils, including all elements of the environment: soils, geology, topography, climate, water resources, flora and fauna, and the effects of anthropogenic activities) of the state of Minas Gerais are in different soil, climate and socio-economics conditions and suitability for the production of agricultural goods is therefore distinct and mapping of agricultural suitability of the state lands is crucial for planning guided sustainability. Geoprocessing uses geographic information treatment techniques and GIS allows to evaluate geographic phenomena and their interrelationships using digital maps. To evaluate the agricultural suitability of state lands, we used soil maps, field knowledge, forest inventories and databases related to Ecological-Economic Zoning (EEZ) of Minas Gerais, to develop a map of land suitability in GIS. To do this, we have combined the maps of soil fertility, water stress, oxygen deficiency, vulnerability to erosion and impediments to mechanization. In terms of geographical expression, the main limiting factor of lands is soil fertility, followed by lack of water, impediments to mechanization and vulnerability to erosion. Regarding agricultural suitability, the group 2 (regular suitability for crops) is the most comprehensive, representing 45.13% of the state. For management levels A and B, low and moderate technological level, respectively, the most expressive suitability class is the regular, followed by the restricted class and last, the adequate class, while for the management level C (high technological level) the predominant class is the restricted. The predominant most intensive use type is for crops, whose area increases substantially with capital investment and technology (management levels B and C).
Ciência e Agrotecnologia 12/2013; 37(6):538-549. DOI:10.1590/S1413-70542013000600007 · 0.64 Impact Factor
[Show abstract][Hide abstract] ABSTRACT: Characterizations of land-cover dynamics are among the most important applications of Earth observation data, providing insights into management, policy and science. Recent progress in remote sensing and associated digital image processing offers unprecedented opportunities to detect changes in land cover more accurately over increasingly large areas, with diminishing costs and processing time. The advent of high-spatial-resolution remote-sensing imagery further provides opportunities to apply change detection with object-based image analysis (OBIA), that is, object-based change detection (OBCD). When compared with the traditional pixel-based change paradigm, OBCD has the ability to improve the identification of changes for the geographic entities found over a given landscape. In this article, we present an overview of the main issues in change detection, followed by the motivations for using OBCD as compared to pixel-based approaches. We also discuss the challenges caused by the use of objects in change detection and provide a conceptual overview of solutions, which are followed by a detailed review of current OBCD algorithms. In particular, OBCD offers unique approaches and methods for exploiting high-spatial-resolution imagery, to capture meaningful detailed change information in a systematic and repeatable manner, corresponding to a wide range of information needs.
International Journal of Remote Sensing 07/2012; 33(14-14):4434-4457. DOI:10.1080/01431161.2011.648285 · 1.65 Impact Factor
[Show abstract][Hide abstract] ABSTRACT: Light Detection and Ranging, or LIDAR, has become an effective ancillary tool to extract forest inventory data and for use in other forest studies. This work was aimed at establishing an effective methodology for using LIDAR for tree count in a stand of Eucalyptus sp. located in southern Bahia state. Information provided includes in-flight gross data processing to final tree count. Intermediate processing steps are of critical importance to the quality of results and include the following stages: organizing point clouds, creating a canopy surface model (CSM) through TIN and IDW interpolation and final automated tree count with a local maximum algorithm with 5 x 5 and 3 x 3 windows. Results were checked against manual tree count using Quickbird images, for verification of accuracy. Tree count using IDW interpolation with a 5x5 window for the count algorithm was found to be accurate to 97.36%. This result demonstrates the effectiveness of the methodology and its use potential for future applications.
[Show abstract][Hide abstract] ABSTRACT: The objective of this study was to evaluate if a multi-resolution segmentation algorithm is sensitive to the RapidEye’s Red Edge band and its benefits for vegetation mapping using GEOBIA and machine learning. We used a high-resolution multi-spectral RapidEye image taken in June, 2010. This image was segmented with a multiresolution segmentation algorithm (MRIS) using a fine scale parameter (300) and thirteen different weights (from ‘0’ to ‘100’) were assigned to the Red Edge spectral band to evaluate its influence in the segmentation and classification process. Each weight generated a segmented image. Attributes related to spectral information, geometry and texture were calculated for each image segment using the eCognition Developer®. Visual interpretation was performed along with field data to select seven classes (Dense vegetation, Meadow, Mining area, Bare land, Rock outcrop, Urban area and Water). A sample of 800 objects described by its attributes was selected from each segmented image. A decision tree approach based on CART was applied to the samples to select the attributes that provides the best separation among the classes
within the scene. An accuracy assessment for the classification using CART was performed to compare the different weights assigned to the Red edge spectral band. Results showed that the Red Edge channel had no significant influence on the segmentation process. The attributes importance rank showed that the index derived from Red Edge channel can be used as input for image classification.
Geographic Object-Based Image Analysis GEOBIA, Rio de Janeiro; 05/2012
[Show abstract][Hide abstract] ABSTRACT: This study analyzed the landscape structure of the Coqueiral Protected Area, located in southern Minas Gerais. We aimed to evaluate the landscape structure in the study area, based on landscape metrics and indicate priority areas for conservation. We use Geographic Information Systems and Remote Sensing tools to construct a land use map from a HCR SPOT 5 satellite image. Landscape structure analysis was carried out through Fragstats software and used landscape metrics. Results showed pasture class was considered as the landscape matrix and occupied almost half of the protected area. Landscape structure analysis showed the landscape is dominated by agropastoral activities. The landscape presented 704 units. Mean patch size area was higher for pasture than semideciduous forest, while semideciduous forest presented higher patch density. Land use classes showed complex shapes indicating higher edge effects. Pasture had the lower patch isolation. Data obtained in this study are relevant for decision making and environmental planning of the Coqueiral Protected Area, allowing suggest priority areas for conservation.
[Show abstract][Hide abstract] ABSTRACT: Conservation Units are among the best methods found to secure biodiversity conservation. The physical and biotic characteristics of high altitude areas such as Serra de Carrancas and Luminárias, in Minas Gerais state, make these places a home to endemic species and to rich biodiversity. However, these environments are highly susceptible to fast advancing erosive processes that potentially lead to soil, habitat and species loss. This study aims to evaluate the physical and biotic characteristics of the Serra de Carrancas and Luminárias region using Natural Vulnerability indicators, and also to propose Conservation Unit implementation in areas which, as per this index, are considered environmentally critical and highly sensitive to anthropic actions. Biotic and abiotic indicators, as managed by a geographic information system, identifi ed the most vulnerable areas in the study site and, given the sensitivity and scope of Serra de Carrancas and Luminárias, a State Park was proposed. The natural vulnerability index proved to be an effective tool to pinpoint prospective conservation unit areas, gathering important environmental factors and thus improving the effi ciency of conservation strategies.
[Show abstract][Hide abstract] ABSTRACT: Multi-temporal images are now of standard use in remote sensing of vegetation during monitoring and classification.Temporal vegetation signatures (i. e., vegetation indices as functions of time) generated, poses many challenges, primarily due tosignal to noise-related issues. This study investigates which methods generate the most appropriate smoothed curves of vegetationsignatures on MODIS NDVI time series. The filtering techniques compared were the HANTS algorithm which is based on Fourieranalyses and Wavelet temporal algorithm which uses the wavelet analysis to generate the smoothed curves. The study was conductedin four different regions of the Minas Gerais State. The smoothed data were used as input data vectors for vegetation classificationby means of artificial neural networks for comparison purpose. A comparison of the results was ultimately discussed in this workshowing encouraging results and similarity between the two filtering techniques used.
[Show abstract][Hide abstract] ABSTRACT: This paper describes the methodology used for monitoring changes in forest and natural vegetation cover in Minas Gerais, Brazil. The state of Minas Gerais is characterized by a large geographical extent and an intense fragmentation history. Both characteristics limit the application of traditional change detection algorithms demanding a large number of trained inpterpreters to separate noise from real land cover changes. Hence, the objective of this work was to develop and implement a procedure for digital change detection which is less sensitive to erros cause by misregistration, phonological state of the vegetation and atmospheric conditions. Landsat images form circa 2005 and circa 2007 acquired over Minas Gerais, as well as a land cover map produced in 2005 were used for change detection. The methodology is based on multiresolution wavelet analysis and local maxima detection in order to pinpoint multscale change objects. The change objects were recursively segmented from coarse to fine scale levels. The results showed that the method is less sensitive to radiometric and geometric misregistration. A new data base relative to vegetation cover classes for the year 2007 was produced and the change statistics were presented for the state of Minas Gerais Palavras-chave: remote sensing, image processing, forest monitoring, sensoriamento remoto, processamento de imagens, monitoramento florestal.
[Show abstract][Hide abstract] ABSTRACT: Temporal vegetation signatures (i. e., vegetation indices as functions of time) generated using the MODIS instrument poses many challenges, primarily due to signal to noise-related issues Bruce et al. (2006). This study investigates which methods best generate smoothed curves of vegetation signatures on MODIS NDVI time series. The filtering techniques compared were the HANTS algorithm, Verhoef (1996), which is based on Fourier analyses and Wavelet temporal algorithm which uses the wavelet analysis to generate the smoothed curves. The smoothed data were used as input data vectors for vegetation classification by means of Artificial neural networks. Statistics of the classifications reveal that the Wavelet filtering algorithm outperforms the original time series and the HANTS fft derived algorithms in all cases in all the classification algorithms.
[Show abstract][Hide abstract] ABSTRACT: O objetivo deste trabalho foi caracterizar a dinâmica sazonal do cerrado, floresta estacional semidecidual e decidual nonorte do estado de Minas Gerais, Brasil. Séries multitemporais dos índices de vegetação NDVI (índice de vegetação da diferençanormalizada) e EVI (índice de vegetação melhorado) derivados do sensor MODIS, foram comparadas analisando o perfil temporale os resultados de classificação das imagens. Os resultados mostraram que: (1) Os índices de vegetação estudados refletiram opadrão sazonal das fisionomias, diferenciando os períodos chuvosos e os períodos de seca; (2) a fisionomia floresta estacionaldecidual apresentou menores valores dos índices e maior variação; (3) as fisionomias cerrado e floresta estacional semidecidualapresentaram alto valores dos índices e baixa variação; (4) de acordo com os resultados das classificações o melhor índice para omapeamento das fisionomias na área de estudo foi o NDVI, porém ambos podem ser usados para avaliar a dinâmica sazonal davegetação; e (5) estudos precisam ser realizados explorando algoritmos de extração de feições para melhorar a acuracidade domapeamento das fisionomias cerrado, floresta decídua e semidecidua na área de estudo.