Urban image classification with semisupervised multiscale cluster kernels

This paper presents a semisupervised support vector machine (SVM) that integrates the information of both labeled and unlabeled pixels efficiently. Method's performance is illustrated in the relevant problem of very high resolution image classification of urban areas. The SVM is trained with the linear combination of two kernels: a base kernel working only with labeled examples is deformed by a likelihood kernel encoding similarities between labeled and unlabeled examples. Results obtained on very high resolution (VHR) multispectral and hyperspectral images show the relevance of the method in the context of urban image classification. Also, its simplicity and the few parameters involved make the method versatile and workable by unexperienced users.

Combining classification techniques to define topo-climatic landscapes

Landscape classification tackles issues related to the representation and analysis of continuous and variable ecological data. In this study, a methodology is created in order to define topo-climatic landscapes (TCL) in the north-west of Catalonia (north-east of the Iberian Peninsula). TCLs relate the ecological behaviour of a landscape in terms of topography, physiognomy and climate, which compound the main drivers of an ecosystem. Selected variables are derived from different sources such as remote sensing and climatic atlas. The proposed methodology combines unsupervised interative cluster classification with a supervised fuzzy classification. As a result, 28 TCLs have been found for the study area which may be differentiated in terms of vegetation physiognomy and vegetation altitudinal range type. Furthermore a hierarchy among TCLs is set, enabling the merging of clusters and allowing for changes of scale. Through the topo-climatic landscape map, managers may identify patches with similar environmental conditions and asses at the same time the uncertainty involved.

Two different 3D biogeochemical models for the NW mediterranean sea: validation with Meris data and intercomparison

Report for the scientific sojourn carried out at the University of New South Wales from February to June the 2007. Two different biogeochemical models are coupled to a three dimensional configuration of the Princeton Ocean Model (POM) for the Northwestern Mediterranean Sea (Ahumada and Cruzado, 2007). The first biogeochemical model (BLANES) is the three-dimensional version of the model described by Bahamon and Cruzado (2003) and computes the nitrogen fluxes through six compartments using semi-empirical descriptions of biological processes. The second biogeochemical model (BIOMEC) is the biomechanical NPZD model described in Baird et al. (2004), which uses a combination of physiological and physical descriptions to quantify the rates of planktonic interactions. Physical descriptions include, for example, the diffusion of nutrients to phytoplankton cells and the encounter rate of predators and prey. The link between physical and biogeochemical processes in both models is expressed by the advection-diffusion of the non-conservative tracers. The similarities in the mathematical formulation of the biogeochemical processes in the two models are exploited to determine the parameter set for the biomechanical model that best fits the parameter set used in the first model. Three years of integration have been carried out for each model to reach the so called perpetual year run for biogeochemical conditions. Outputs from both models are averaged monthly and then compared to remote sensing images obtained from sensor MERIS for chlorophyll.

Testing low-altitude infrared digital photography from a mini-UAV to retrieve information for biological conservation

Satellite remote sensing imagery is used for forestry, conservation and environmental applications, but insufficient spatial resolution, and, in particular, unavailability of images at the precise timing required for a given application, often prevent achieving a fully operational stage. Airborne remote sensing has the advantage of custom-tuned sensors, resolution and timing, but its price prevents using it as a routine technique for the mentioned fields. Some Unmanned Aerial Vehicles might provide a “third way” solution as low-cost techniques for acquiring remotely sensed information, under close control of the end-user, albeit at the expense of lower quality instrumentation and instability. This report evaluates a light remote sensing system based on a remotely-controlled mini-UAV (ATMOS-3) equipped with a color infra-red camera (VEGCAM-1) designed and operated by CATUAV. We conducted a testing mission over a Mediterranean landscape dominated by an evergreen woodland of Aleppo pine (Pinus halepensis) and (Holm) oak (Quercus ilex) in the Montseny National Park (Catalonia, NE Spain). We took advantage of state-of-the-art ortho-rectified digital aerial imagery (acquired by the Institut Cartogràfic de Catalunya over the area during the previous year) and used it as quality reference. In particular, we paid attention to: 1) Operationality of flight and image acquisition according to a previously defined plan; 2) Radiometric and geometric quality of the images; and 3) Operational use of the images in the context of applications. We conclude that the system has achieved an operational stage regarding flight activities, although with meteorological limits set by wind speed and turbulence. Appropriate landing areas can be sometimes limiting also, but the system is able to land on small and relatively rough terrains such as patches of grassland or short matorral, and we have operated the UAV as far as 7 km from the control unit. Radiometric quality is sufficient for interactive analysis, but probably insufficient for automated processing. A forthcoming camera is supposed to greatly improve radiometric quality and consistency. Conventional GPS positioning through time synchronization provides coarse orientation of the images, with no roll information.

Application of a roughness-length representation to parameterize energy loss in 3-D numerical simulations of large rivers

Recent technological advances in remote sensing have enabled investigation of the morphodynamics and hydrodynamics of large rivers. However, measuring topography and flow in these very large rivers is time consuming and thus often constrains the spatial resolution and reach-length scales that can be monitored. Similar constraints exist for computational fluid dynamics (CFD) studies of large rivers, requiring maximization of mesh-or grid-cell dimensions and implying a reduction in the representation of bedform-roughness elements that are of the order of a model grid cell or less, even if they are represented in available topographic data. These ``subgrid'' elements must be parameterized, and this paper applies and considers the impact of roughness-length treatments that include the effect of bed roughness due to ``unmeasured'' topography. CFD predictions were found to be sensitive to the roughness-length specification. Model optimization was based on acoustic Doppler current profiler measurements and estimates of the water surface slope for a variety of roughness lengths. This proved difficult as the metrics used to assess optimal model performance diverged due to the effects of large bedforms that are not well parameterized in roughness-length treatments. However, the general spatial flow patterns are effectively predicted by the model. Changes in roughness length were shown to have a major impact upon flow routing at the channel scale. The results also indicate an absence of secondary flow circulation cells in the reached studied, and suggest simpler two-dimensional models may have great utility in the investigation of flow within large rivers. Citation: Sandbach, S. D. et al. (2012), Application of a roughness-length representation to parameterize energy loss in 3-D numerical simulations of large rivers, Water Resour. Res., 48, W12501, doi: 10.1029/2011WR011284.

Textural classification of land cover using support vector machines: an empirical comparison with parametric, non parametric and hybrid classifiers in the Bolivian Amazon

Land cover classification is a key research field in remote sensing and land change science as thematic maps derived from remotely sensed data have become the basis for analyzing many socio-ecological issues. However, land cover classification remains a difficult task and it is especially challenging in heterogeneous tropical landscapes where nonetheless such maps are of great importance. The present study aims to establish an efficient classification approach to accurately map all broad land cover classes in a large, heterogeneous tropical area of Bolivia, as a basis for further studies (e.g., land cover-land use change). Specifically, we compare the performance of parametric (maximum likelihood), non-parametric (k-nearest neighbour and four different support vector machines - SVM), and hybrid classifiers, using both hard and soft (fuzzy) accuracy assessments. In addition, we test whether the inclusion of a textural index (homogeneity) in the classifications improves their performance. We classified Landsat imagery for two dates corresponding to dry and wet seasons and found that non-parametric, and particularly SVM classifiers, outperformed both parametric and hybrid classifiers. We also found that the use of the homogeneity index along with reflectance bands significantly increased the overall accuracy of all the classifications, but particularly of SVM algorithms. We observed that improvements in producer’s and user’s accuracies through the inclusion of the homogeneity index were different depending on land cover classes. Earlygrowth/degraded forests, pastures, grasslands and savanna were the classes most improved, especially with the SVM radial basis function and SVM sigmoid classifiers, though with both classifiers all land cover classes were mapped with producer’s and user’s accuracies of around 90%. Our approach seems very well suited to accurately map land cover in tropical regions, thus having the potential to contribute to conservation initiatives, climate change mitigation schemes such as REDD+, and rural development policies.

Active Learning of Very-High Resolution Optical Imagery with SVM: Entropy vs Margin Sampling

An active learning method is proposed for the semi-automatic selection of training sets in remote sensing image classification. The method adds iteratively to the current training set the unlabeled pixels for which the prediction of an ensemble of classifiers based on bagged training sets show maximum entropy. This way, the algorithm selects the pixels that are the most uncertain and that will improve the model if added in the training set. The user is asked to label such pixels at each iteration. Experiments using support vector machines (SVM) on an 8 classes QuickBird image show the excellent performances of the methods, that equals accuracies of both a model trained with ten times more pixels and a model whose training set has been built using a state-of-the-art SVM specific active learning method

Machine learning algorithms for spatial data. Case studies: Environmental pollution, natural hazards, renewable resources

This paper presents general problems and approaches for the spatial data analysis using machine learning algorithms. Machine learning is a very powerful approach to adaptive data analysis, modelling and visualisation. The key feature of the machine learning algorithms is that they learn from empirical data and can be used in cases when the modelled environmental phenomena are hidden, nonlinear, noisy and highly variable in space and in time. Most of the machines learning algorithms are universal and adaptive modelling tools developed to solve basic problems of learning from data: classification/pattern recognition, regression/mapping and probability density modelling. In the present report some of the widely used machine learning algorithms, namely artificial neural networks (ANN) of different architectures and Support Vector Machines (SVM), are adapted to the problems of the analysis and modelling of geo-spatial data. Machine learning algorithms have an important advantage over traditional models of spatial statistics when problems are considered in a high dimensional geo-feature spaces, when the dimension of space exceeds 5. Such features are usually generated, for example, from digital elevation models, remote sensing images, etc. An important extension of models concerns considering of real space constrains like geomorphology, networks, and other natural structures. Recent developments in semi-supervised learning can improve modelling of environmental phenomena taking into account on geo-manifolds. An important part of the study deals with the analysis of relevant variables and models' inputs. This problem is approached by using different feature selection/feature extraction nonlinear tools. To demonstrate the application of machine learning algorithms several interesting case studies are considered: digital soil mapping using SVM, automatic mapping of soil and water system pollution using ANN; natural hazards risk analysis (avalanches, landslides), assessments of renewable resources (wind fields) with SVM and ANN models, etc. The dimensionality of spaces considered varies from 2 to more than 30. Figures 1, 2, 3 demonstrate some results of the studies and their outputs. Finally, the results of environmental mapping are discussed and compared with traditional models of geostatistics.

Phlebotominae spatial distribution asssociated with a focus of tegumentary leishmaniasis in Las Lomitas, Formosa, Argentina, 2002

Las Lomitas, Formosa, Argentina, reported 96 cases of tegumentary leishmaniasis during 2002. The urban transmission was suggested although previous outbreaks were related with floods of the Bermejo river (BR) 50 km from the village. Phlebotomine collections were performed during March 2002 to define the spatial distribution of risk, together with satellite imaginery. The phlebotomine/trap obtained was 1679.5 in the southern BR shore, 1.1 in the periruban-rural environment and 2.3 in the northern Pilcomayo river marshes. Lutzomyia neivai was the prevalent species (91.1%) among the 2393 phlebotomine captured, and it was only found in the BR traps. The other species were L. migonei (7.9%), L. cortelezzii (0.9%), and Brumptomyia guimaraesi (0.1%). The satellite images analysis indicates that the fishing spots at the BR were significantlyoverflowed during the transmission peak, consistent with fishermen recollections. This spatial restricted flood might concentrate vectors, reservoirs, and humans in high places. Therefore, both the spatial distribution of vectors and the sensor remoting data suggests that in Las Lomitas area the higher transmission risk it is still related with the gallery forest of the BR, despite of the urban residence of the cases. The surveillance and control implications of these results are discussed.

Cluster kernels for semisupervised classification of VHR urban images

In this paper, we present and apply a semisupervised support vector machine based on cluster kernels for the problem of very high resolution image classification. In the proposed setting, a base kernel working with labeled samples only is deformed by a likelihood kernel encoding similarities between unlabeled examples. The resulting kernel is used to train a standard support vector machine (SVM) classifier. Experiments carried out on very high resolution (VHR) multispectral and hyperspectral images using very few labeled examples show the relevancy of the method in the context of urban image classification. Its simplicity and the small number of parameters involved make it versatile and workable by unexperimented users.

Towards Detecting Changes in Underwater Image Sequences

Detecting changes between images of the same scene taken at different times is of great interest for monitoring and understanding the environment. It is widely used for on-land application but suffers from different constraints. Unfortunately, Change detection algorithms require highly accurate geometric and photometric registration. This requirement has precluded their use in underwater imagery in the past. In this paper, the change detection techniques available nowadays for on-land application were analyzed and a method to automatically detect the changes in sequences of underwater images is proposed. Target application scenarios are habitat restoration sites, or area monitoring after sudden impacts from hurricanes or ship groundings. The method is based on the creation of a 3D terrain model from one image sequence over an area of interest. This model allows for synthesizing textured views that correspond to the same viewpoints of a second image sequence. The generated views are photometrically matched and corrected against the corresponding frames from the second sequence. Standard change detection techniques are then applied to find areas of difference. Additionally, the paper shows that it is possible to detect false positives, resulting from non-rigid objects, by applying the same change detection method to the first sequence exclusively. The developed method was able to correctly find the changes between two challenging sequences of images from a coral reef taken one year apart and acquired with two different cameras

Large-Area Photo-Mosaics Using Global Alignment and Navigation Data

Seafloor imagery is a rich source of data for the study of biological and geological processes. Among several applications, still images of the ocean floor can be used to build image composites referred to as photo-mosaics. Photo-mosaics provide a wide-area visual representation of the benthos, and enable applications as diverse as geological surveys, mapping and detection of temporal changes in the morphology of biodiversity. We present an approach for creating globally aligned photo-mosaics using 3D position estimates provided by navigation sensors available in deep water surveys. Without image registration, such navigation data does not provide enough accuracy to produce useful composite images. Results from a challenging data set of the Lucky Strike vent field at the Mid Atlantic Ridge are reported

A geoprocessing approach for studying and controlling schistosomiasis in the state of Minas Gerais, Brazil

Geographical information systems (GIS) are tools that have been recently tested for improving our understanding of the spatial distribution of disease. The objective of this paper was to further develop the GIS technology to model and control schistosomiasis using environmental, social, biological and remote-sensing variables. A final regression model (R² = 0.39) was established, after a variable selection phase, with a set of spatial variables including the presence or absence of Biomphalaria glabrata, winter enhanced vegetation index, summer minimum temperature and percentage of houses with water coming from a spring or well. A regional model was also developed by splitting the state of Minas Gerais (MG) into four regions and establishing a linear regression model for each of the four regions: 1 (R² = 0.97), 2 (R² = 0.60), 3 (R² = 0.63) and 4 (R² = 0.76). Based on these models, a schistosomiasis risk map was built for MG. In this paper, geostatistics was also used to make inferences about the presence of Biomphalaria spp. The result was a map of species and risk areas. The obtained risk map permits the association of uncertainties, which can be used to qualify the inferences and it can be thought of as an auxiliary tool for public health strategies.