Virtual reality systems have been extending their usability in many areas that could not be contemplated before. One of these fields is paleontology, which can now use virtual reality to build realistic models of real fossils with many goals, including preservation of originals, optimized visualization, or even for restoring missing parts. In this paper we present an immersive system where a entire scenario is reconstructed using digital photogrammetry and Mosis LAB application. The system has shown useful for many applications from touristic purposes and paleontological studies, since the immersive system provides an increased sense of manipulation, as well as the possibility of detailed inspection both inside and outside structures.

The Mean Tropospheric Temperature (Tm) is important for the sake of the conversion between Zenith Path Delay (ZPD) to Precipitable Water Vapor (PWV) because of the proportionality constant that takes Tm to be calculated. In Brazil, Tm modeling’s been traditionally performed using Multiple Linear Regression Models (MLR). However, recent works suggest the use of Deep Learning methods to model Tm values. In this work, we propose models based on Convolutional Neural Networks (CNN) using radiosonde data from 1961 to 2010. The results show that CNN models can outperform the traditional methods considering different statistic metrics like R, standard deviation, and RMSE.

Most fracture properties, such as orientation and density, are acquired by interpreting data obtained at the wellbores, while fracture properties between wells are typically derived from seismic data. However, this information is sparse or has low resolution leading to the study and analysis of outcrops. The data acquisition in outcrops is facilitated by its 3D representation in Digital Outcrop Models (DOM) obtained from LiDAR and Photogrammetry. It also allows virtual interpretation and fracture detection. In either case, the 3D fracture data need to be clustered in fracture sets to allow the statistical analysis necessary to upscale and resample the DFN. This clusterization is carried out by methods like k-means and fuzzy, however with caveats, like the prior definition of the number of clusters. In this work, we propose the use of agglomerative hierarchical clustering to cluster 3D fracture data obtained by a KD-Tree segmentation algorithm. Results showed that when compared to k-means, the proposed method presented more compact clusters and balance when considering Fisher’s statistics of dispersion.

Hyperspectral images often have low spatial resolution due to the sensor sizes required to capture the required spectral response. Super-resolution (SR) techniques try to mitigate this by injecting more detail in the upscaled image, either with numerical methods or deep learning and Convolution Neural Networks. In the deep learning methods, the models learn image details by inferring a high-resolution (HR) image from a synthetic low-resolution (LR) image that simulates the natural degradation of sensors by applying resampling (to reduce the image detail) and noising (to add small errors and interference). Often disregarded in the literature, the resampling method applied to generate the synthetic image can impact greatly the deep learning model training. This work, evaluate several resampling techniques to measure this impact using the Harvard hyperspectral dataset. Results showed that the Lanczos filter was the best among eight other resampling methods. The Nemenyi and Friedman ranking statistical tests also indicated that the Cubic-Spline, Bicubic, and RMS achieved good results.

Improvement in spatial resolution of remote sensing images will always be a hot-topic since the sharpening of image objects is mandatory for many appications. Even thought image instruments have received great improvements in the recent years, super-resolution techniques are welcome to increase even more the level of quality of the data for further interpretations. In this paper we present a multi-frame super-resolution approach that uses a set of UAV images acquired at the same spot, but with slight different perspectives caused by random fluctuations. The small off-sets between consecutive pixels of the low spatial resolution images are considered at subpixel level to feed an arithmetic system of equations able to produce high resolution pixels, which are the unknowns of the system. The results are soundness and could show visual enhancement. However, further developments is need to in deep undertanding and possible advancement of the designed approach.

Carbonate outcrops are widely studied due to their character-istic analogous to oil and gas reservoirs. Analyzes made with data from these outcrops are used to assist in the description of reservoirs located at great depths. These rocky expo-sures can be easily detected in satellite images using machine learning. Considering the diversity of techniques used in the identification/classification of features on the satellite image data, this work aimed to evaluate the best machine learning technique for identifying carbonate outcrops in Landsat 8 data, considering the carbonate outcrops from the Jandaíra formation, in northeast Brazil. We employed techniques like Grid Search, cross-validation, and validation statistics like Friedman and Nemenyi. Considering the metrics F1 score and Mathews Correlation Coefficient, the MLP technique presented the best results, achieving a F1 score of 0.823, against the SVM with a F1 score of 0.810 and the Random Forest technique with a F1 score of 0.812.

Differently form RGB photos, multi-frame acquisitions of hyperspectral data are unfeasible due to the pushbroom mechanisms of hyperspectral sensors. Therefore, 3D digi-tal surface models including good quality hyperspectral in-formation are available only by integrating 2D images to the point cloud generated using photogrametric cameras or laser scanners. However, geometric differences presented by sen-sor mechanisms resulting in varied perspectives of the same objects present on the scene deeply harm the spatial registration, and thus, the post integration procedure. In this paper we suggest a pre-processing step aiming to adjust objects to the same geometric perspective in order to later integrate the RGB cloud mesh to the hyperspectral data in a more efficient manner. We firstly build the RGB point cloud for later simu-lating the same perspective of push-broom sensor generating a RGB orthomosaic similar to the hyperspectral image. Pre-liminary results indicate an increasing in the number of tie points when registering is performed using the resulting syn-thetic sampling compared to the registration of original data.

Kerogen type in source rocks is directly related to its hydrocarbon generation potential. Its determination is often carried out with destructive methods. This study presents a non-destructive technique as an alternative to determine kerogen type using hyperspectral data and machine learning techniques. To present the technique, models were training using Support Vector Machines, K Nearest Neighbors, and Random Forest classifiers on spectral data collected in rock samples acquired from Taubaté Basin, Brazil, of an outcrop with high hydrocarbon generation potential. The models were trained and evaluated using spectral signatures measured with a spectroradiometer and the results were also tested on hyperspectral images of the samples. The experiments described here achieved accuracy above 0.8 with precision and recall above 0.62 and 0.8, respectively, for every kerogen type, indicating the soundness of the classification.

Fractures affect the intrinsic properties of permeability and porosity of reservoir geobodies, making its network characterization an important task for fluid flow modeling. Direct acquisition of data on reservoirs is labor-intensive and generally produces sparse information. Thus, the study of analogue outcrops with similar characteristics is often carried out by using unmanned aerial vehicle image acquisition and digital photogrammetry. However, the accurate automatic recognition of the fractures network over the outcrop images remains a challenge. Image segmentation methods based on convolution neural networks (CNNs) were successfully applied in medicine, biology, and other areas, however, not yet in geological fracture detection. This work proposes the validation of two popular CNNs - Segnet and Unet - for pixel-to-pixel segmentation targeting fracture detection. Initial results showed acceptable scores of the metrics mean intersection over union (mIoU) and dice intersection (F1) in both CNNs.

Very high spatial resolution data seems to reach its maximum for orbital images due to unavoidable atmospheric interactions. At the same time, special hyperspectral cameras are being developed to operate on-board manned or unmanned aircrafts at a fixed optics, which prevents its using for imaging near objects in laboratory conditions. Both limitations can only be surpassed by using super-resolution principles. In this paper, we present a multi-looking approach for enhancing the spatial resolution of images acquired by systems that exhausted their natural ability to provide finer images. The method exploits multiple image takes with controlled spatial differences to produce a higher resolution output. Experiments with static hyperspectral sensor and synthetic data have proven the approach is sound and robust to many applications (e.g., rock samples).

New sensors aboard recently launched satellites have induced the development of several measures aimed to indicate the presence of many materials over the Earth. Karsts are places rich in carbonate rocks and present large economic and environmental importance. This paper aimed at assessing the performance and consistency of different carbonate estimators derived from orbital images acquired over a controlled karst area. Experiments were assisted by a multi-scaled reference data built through a high spatial resolution Unmanned Aerial Vehicle (UAV) image acquired over the selected area. Results show a considerable unconformity among selected measures and better performance presented by indices exploiting measures along visible and infrared spectral regions.

This paper presents a supervised classification approach based on the analysis of color tones depicted by objects in scenes with natural targets. In some environmental classification problems, the existing classes occur in subtle color variations, leading to a limited and inefficient set of samples collected by the user. In the proposed framework, a reduced Hue-Saturation-Value (HSV) color system is used to select appropriate shades of color standing for classes of interest in a supervised fashion. The classification strategy is designed to emulate decisions made by the analyst providing high level of generalization in problems where classes occur in many color tones of similar hue. An experiment performed with Amazon rainforest scene including deforestation activities was classified using the proposed technique and several well-known classification approaches. The method proved to be useful for noncomplex problems, also providing simplicity, intuitiveness, and the ability to generalize the training process.

This study describes the application of an automated methodology for retroactive detection of Amazon deforestation in Rondônia-Brazil. The system officially adopted by the Brazilian government to detect the deforestation fragments was implemented in 1988. However, the detailed mapping was not possible, since only annual rates were counted, but with no spatial description. Only in 2000, the system started to provide data on digital formats, based on an analog verification performed by visual interpretation, which was able to produce detailed deforestation mapping. The automated detection used in this work reduces user interference, as well as the time spent in the analysis, providing objectivity to the mapping. By adding the available data to the data produced by proposed method, the information on deforestation occurred before 2000, which reached 41456,084 km2of deforested area, was completed and a database was provided detailing three decades of research in the Amazon rainforest.

This paper presents a new paradigm for object based classification of multispectral images. Instead of classifying objects only after the segmentation process is completed, it is proposed to intercept the early stages of the segmentation by iteratively performing classification tests to under growing regions. By applying this simultaneous analysis, mislabeling of objects considered only after segmentation is completely done can be avoided. The proposed technique assumes that some growing regions can present higher membership to a particular class when comparing to the final object in which it is included. A Bayesian framework was applied in classification tests performed by pixel based, traditional object based, and the proposed technique were performed. The results show the soundness of the proposed method when comparing overall accuracies with a reference map.

In this paper, we present a novel domain adaptation technique aimed at providing reliable change detection maps for a series of image pairs acquired on the same area at different times. The proposed technique exploits the polar change vector analysis method and assumes that the reference data for characterizing a specific change of interest are available only for a pair of images (source domain). Then, it exploits the knowledge learned from the source domain and adapts it to other pairs of images belonging to the time series (target domains) to be analyzed. The proposed technique is able to handle possible radiometric differences among images adapting in an unsupervised way the decision rule estimated on the source domain to the target domains through variables estimated directly on the target images. The proposed approach has been applied to two data sets made up of time series of Landsat Thematic Mapper images. In one case, the change of interest is related to evolution of deforestation, while in the other case, it is related to burned area detection. Experimental results show the effectiveness of the proposed technique.

This paper aims to use unique features of hyperspectral data on an automatic process for outlining individual tree crowns (ITCs) in a tropical forest area, with special focus on semi-deciduous species. In order to enhance biophysical and biochemical properties of canopy species, a set of vegetation indices were computed. These indices served as input for a region growing segmentation algorithm that takes into account mutual similarity of pixels and spectral separability between neighbor segments. Segmentation output was evaluated on the basis of a score computed with the proportion of the area of the segments located within manually delineated ITCs. Results show that the segmentation approach is able to automatically delineate up to 70% of the control ITCs.

This paper presents an adaptive framework for detection of changes of relevance occurring in image time series in a recursive way. With the availability of reference data for only one image pair from the time series (source domain), the proposed methodology employs change vector analysis in the 3-dimensional spherical domain to determine a decision region R associated with the change of relevance. Then, by exploiting the similarity among domains, the same kind of change can be detected by adapting R to the rest of image pairs belonging to the time series. The methodology was tested in a multispectral time series made up by TM-Landsat images marked by sequential deforestation activities in the Amazon with reference data. The quantitative analysis of the results indicates the soundness of the proposed approach.

This paper presents an alternative object based classification for multispectral remote sensing images. Instead of classifying the images after the segmentation process, it is suggested to involve some steps of objects recognition during the segmentation process in order to improve the final classification results. The methodology is based on the statistical distribution of object classes. Experiments were performed with a TM-Landsat image and the results were compared with a reference data. The results indicate the soundness of the proposed methodology.

A problem that is frequently arising in the spectral mixture analysis is how to correctly identify the endmembers present in the scene. In the analysis of image data covering natural scenes, vegetation, bare soil, and shade/water are commonly assumed as endmembers, but other endmembers may also be present. This paper investigates an approach based on the analysis of residuals produced by the linear spectral mixing model for identifying potential missing endmembers. The basic proposition consists in assuming that larger residuals are caused by missing endmembers. The image is segmented in terms of the residuals, and the Kolmogorov-Smirnov test is applied to group segments that show similar residuals and are thus likely to include the same missing endmember. An approach to estimate the spectral response of the missing endmembers is also investigated. The proposed methodology is tested by using Thematic Mapper Landsat and Coupled Charge Device China-Brazil Earth Resources Satellite image data. In addition to vegetation, bare soil, and shade/water, two additional endmembers were included as missing endmembers (clouds and water bodies with a large load of suspended sediments). The tests have shown that the proposed methodology is capable of detecting image regions that include missing endmembers and of correctly estimating the corresponding spectral responses.

In this paper, a new concept to change detection in remote sensing multitemporal images is presented. Traditional methods are generally concerned to label pixels into two exhaustive classes: change or no change. Even this approach is more common used, real environmental changes tend to occur in a continuum, rather than sudden manner. The proposed methodology is based on Bayesian framework and fraction images in order to classify pixels according to degrees of membership to the class change, in a fuzzy-like fashion. An experiment is performed employing synthetic image simulating realistic changes. The result shows that the methodology can adequately tell about the gradual changes occurred between two dates.

In this paper, a methodology for automatic detection of burned areas is suggested. The classification criterion is performed using Bayesian statistical parameter (mean and covariance matrix) extracted automatically using the Expectation Maximization algorithm and taking into account the spectral similarity between burned and flooded areas. In this work the final process involves the application of morphological operators of erosion and dilation of images in order to insert information from the spatial context, refining the final map. Experiments were conducted to a TM-Landsat scene with areas affected by fires and seasonal flooding. The results show that the accuracy is increased with the consideration of flooding mask and the subsequent application of spatial context, reaching values up to 97% of accuracy when compared with a reference map.

This paper proposes an analysis on the residual term resulting from the Linear Spectral Mixing Model (SLMM) solution in order to access model uncertainty. The framework employed here is based on analysis of data produced initially by unmixing of vegetation, bare soil and shade/water, whose are commonly used as standard endmembers. We suggest procedures to identify missing components in the mixture problem and automatically compute the spectral endmember values for these components directly from image data and residual information. The techniques proposed have been tested on real TM-Landsat. The results obtained promises and confirm the validity of the proposed approach.