This research investigates the application of iterative point-based sparse annotations for semantic segmentation of cars in remote-sensing imagery, to mitigate the challenges associated with laborious and expensive data labeling processes. Car labeling considered the selection of point shapefiles in the Geographic Information System with a value of 1 for the specific class of car, 0 to background and a value of -1 outside the intended target. The semantic segmentation model was the U-Net architecture, with an Efficient-net-B7 backbone and a modified cross-entropy loss function. The experimental evaluation uses the BSB Vehicle Dataset, encompassing two classes (background and vehicles). The results showcase promising improvements, particularly in error-prone classes, as more samples are iteratively added during training. This approach presents a viable and time-efficient alternative for dataset creation, leveraging sparse annotations that are incrementally enhanced. Our pipeline included five iterative rounds in which the IoU increased more than 30% from the first round to the fifth, achieving 60% IoU with 0.059% of the total annotated data. Showing a very good performance with less than 0.1% of pixels annotated. This research advances the field by proposing a rapid and cost-effective method for generating high-quality datasets in remote sensing.

This study addresses the challenge of amodal segmentation in computer vision, a change in basic assumptions towards perceiving objects holistically, even when partially occluded, deviating from the traditional modal perspective that predominantly focuses on visible elements. Thus, we propose a new approach for the amodal segmentation of top-view aerial images, with particular attention to the first layer of elements, constituted by asphalt and natural soils, normally occluded by different objects (trees, buildings, and vehicles). This proposed methodology is data-centric, assigning weights to specific image sections and distinguishing non-visible elements. The best model used the U-Net architecture with Efficient-net-B7 as the backbone and can accurately classify occluded segments, achieving an Intersection over Union (IoU) greater than 80% for most classes. The developed method provides a basis for exploring amodal segmentation based on data-centric models, impacting our understanding of complex and occlusion-prone environments, such as urban environments.

This study investigates the application of iterative sparse annotations for semantic segmentation in remote-sensing imagery, focusing on minimizing the laborious and expensive data labeling process. By leveraging Geographic Information Systems (GIS), we implemented circular polygon shapefiles to label portions of each class, attributing a value of -1 outside these polygons. The model training used the simplified BSB Aerial Dataset with eight classes. The semantic segmentation model was U-Net architecture with the Efficient-net-B7 backbone and a modified cross-entropy loss function. Our results showed promising improvement, particularly in error-prone classes, with the iterative addition of more samples. This approach suggests a quicker method for dataset creation using sparse, iteratively enhanced annotations. Future work will aim to implement further iterative rounds to approximate the results of continuous labeling, thereby enhancing the efficiency of semantic segmentation in large-scale remote-sensing images.

This research aims to investigate the use of semantic segmentation and Shuttle Radar Topography Mission (SRTM) data in detecting natural karst depressions developed on the carbonate rocks of the Neoproterozoic Bambuí Group in Western Bahia, Brazil. The study area is a karst landscape containing depressions enclosed in limestone, many forming lakes. The methodology had the following steps: (a) visual interpretation of karst depressions from Sentinel-2 and OLI-Landsat 8 images; (b) generation of DEM-based sink depth plus nine morphometric attributes; (c) selection of 128x128-pixel samples for training (1600), validation (400), and testing (400) considering two channels (DEM and sink depth based on DEM) and eleven channels (the two previous ones and the morphometric attributes); and (d) semantic segmentation using U-Net architecture with EfficientNet-B7 backbone. The accuracy metrics were 98.26, 72.82, 79.50, 79.16, and 65.51 for OA, precision, recall, F-score, and IoU when considering SRTM plus morphometric attributes (11 channels).

Panoptic segmentation combines instance and semantic seg-mentation, enabling the classification of objects and back-grounds. It is still a method little explored in the remote sensing field, mostly due to the difficulty of generating the data. Moreover, the beach areas have great interest due to many objects and elements that may guide public policies. In this regard, we propose the first study on beach areas using panop-tic segmentation and the first panoptic segmentation study using multispectral data. We used the Gram-Schmidt pan-sharpening method for the multispectral bands and created a dataset with 850 samples with 128x 128 dimensions in the COCO panoptic annotation format. To evaluate the dataset, the Panoptic-FPN was used with modifications in the input (changing from three to eight channels). Results show 59.43 Panoptic Quality (PQ), 77.96 Segmentation Quality (SQ), and 75.07 Recognition Quality (RQ).

2D representations of 3D scenes generate occlusions among different targets. Understanding targets by only seeing parts of them is referred to as an amodal perception, which is still unexplored in remote sensing. Thus, we propose integrating this concept using peculiarities of remote sensing Nadir images to classify non-visible targets at a pixel level. Nadir images present a hierarchical order of occlusions, allowing us to separate different layers. We developed a dataset with 600 images and three classes (roads, vehicles, and trees) with in-dependent labelling for each class. Any semantic segmentation model is suitable for this task, but we explored the U-net architecture with three backbones (Efficient-net-B7, ResNet-101, and ResNeXt-101). The evaluation considered the IoU metric, providing 80% for the best model (Efficient-net-B7). Future studies aim to extend this approach by introducing competing classes among each layer and increasing the number of samples and categories.

This letter proposes a novel method to obtain panoptic predictions by extending the semantic segmentation task with a few non-learning image processing steps, presenting the following benefits: 1) annotations do not require a specific format [e.g., common objects in context (COCO)]; 2) fewer parameters (e.g., single loss function and no need for object detection parameters); and 3) a more straightforward sliding windows implementation for large image classification (still unexplored for panoptic segmentation). Semantic segmentation models do not individualize touching objects, as their predictions can merge; i.e., a single polygon represents many targets. Our method overcomes this problem by isolating the objects using borders on the polygons that may merge. The data preparation requires generating a one-pixel border, and for unique object identification, we create a list with the isolated polygons, attribute a different value to each one, and use the expanding border (EB) algorithm for those with borders. Although any semantic segmentation model applies, we used the U-Net with three backbones (EfficientNet-B5, EfficientNet-B3, and EfficientNet-B0). The results show that the following hold: 1) the EfficientNet-B5 had the best results with 70% mean intersection over union (mIoU); 2) the EB algorithm presented better results for better models; 3) the panoptic metrics show a high capability of identifying things and stuff with 65 panoptic quality (PQ); and 4) the sliding windows on a$2560\times 2560$-pixel area has shown promising results, in which the ratio of merged objects by correct predictions was lower than 1% for all classes.

Center pivots are a modern irrigation technique mainly applied in precision agriculture, once it has high efficiency in water consumption and low labor workers when compared to traditional irrigation methods. Knowing their location is valuable since monitoring, evaluating, and estimating essential features in the lands becomes easier, remote sensing is a robust tool to act upon this kind of problem. To identify center pivots, we used a deep residual U-Net with a pixel comparison at image reconstruction to enhance results. We obtained a validation loss of 0.19, which adds up with pixel comparison. Results were satisfactory, with 2070 correct identifications from a total of 2109 center pivots (98.15%). Future studies to improve these results would require more data in different places and seasons.

Time series of satellite images allows better monitoring and detecting the dynamics of urban growth. The objective of this research is to detect the urban area between the city of Rio de Janeiro and São Paulo in the period of 2014-2015 using MODIS digital time series. We used the product MOD13Q1 referring to the Normalized Difference Vegetation Index (NDVI) 16-day composite data, with spatial resolution of 250 meters. The high variety of elements imposes a great difficulty in mapping urban areas. Therefore, we calculated the nearest neighbor using the Euclidian distance for each time signature. The different image group of the urban targets were into a single image considering the minimum value of each pixel within the set. Therefore, a limit value separated the urban areas from the rest. This methodology allowed the detection of urban areas considering their diversity. The algorithm is written in C ++ language.

The present research aims to evaluate the temporal dynamics of the normalized difference vegetation index (NDVI) by the exclusion of fire in the Águas Emendadas Ecological Station, Central Brazil. The study area presents savanna vegetation, in which fire is an integral part of the ecosystem. The results indicate that the fire exclusion for more than 15 years resulted in a significant increase of NDVI within the ecological season. This increase in vegetation may result in more severe fires in the future given the greater availability of fuels.

Mathematical modeling is being increasingly used to predict events occurring in nature. Within the diverse existing models, one which stands out is the SHALSTAB. This model of prediction of shallow landslide occurrence was applied in Parque Nacional da Serra dos Órgãos (PARNASO) by using data for average monthly pluviosity, aiming to identify, within the landscape, the spatial variability at places prone to shallow landslide throughout the year. The methodology is sectioned into the following stages: (a) elaboration of the digital elevation model (DEM) and its derived maps, such as slope and contribution area, (b) application of the SHALSTAB model, considering the various events of rainfall throughout the year, and (c) quantification of areas prone to landslide for each rainfall event occurred. The model results indicate the dynamics of the locations which present instability due to the seasonality of rainfall intensity.

Natural disasters are one of the world’s greatest socioeconomic problems and one of the most notable of which is that of mass movements, which are natural phenomena that change the relief and cause great damage to mankind. Mass movements can be classified by: type of material involved, velocity and mechanism of movement, type of deformation, geometry of the displaced mass and water content. Among them, the shallow landslides stand out. These movements are triggered during intense rainfall, frequently in the summer. Shallow landslides transform into debris flow mainly due to water infiltration. They reach long distances, velocity and high transport capacity, including large boulders. It is essential not only to predict shallow translational landslides but also to predict debris flow occurrence and runout as a secondary effect. The aim of this study is to develop a methodology which combines a model to predict landslides with another one that determines debris flows pathways and depositions. The study area is located in the west slope of Maciço da Tijuca, involving the Quitite and Papagaio river basins, in the Jacarepaguá neighbourhood of Rio de Janeiro, Brazil.

Digital change detection is the computerized process of identifying changes in the state of an object, or other earth-surface features, between different dates. During the last years, a large number of change detection methods have evolved that differ widely in refinement, robustness and complexity. This study aims to develop a program to detect and delineate landscape changes automatically over multiple scales using a formulation of spectral classifiers. The procedure calculates, for each pixel, SAM, SCM or Euclidian distance value between the spectra at time t1 and t2. By considering a high threshold value, it is possible to define points with the same spectral behavior and probably without alteration during the period. In particular, this method allows for the automatic identification of invariant points to calibrate remote-sensing images, without visual interpretation data. Users' program establishes the spectral change detection (SCD) method (SAM-SCD, SCM- SCD and Euclidian Distance-SCD). Thus, the program allows to work simultaneously with the collection of temporal images. This realization is an important task in landscape analysis and remote sensing.

A consistent data acquisition interval among multi-temporal images is necessary to accurate landscape change detection and temporal-series signatures. Orbital images are difficult to maintain a temporal precision due to the different interferences that generate missing data. The correct handling of missing data is a difficult problem in data analysis and often depends on your specific situation. This missing information can be replaced using an interpolation method. In this paper is proposed a new algorithm that interpolates multitemporal images. This new computation method uses the cubic-spline interpolation technique to trace the reflectance and NDVI behaviors along time. The performance of the cubic-spline interpolation technique in the determination of NDVI temporal series is verified in terms of accuracy.

Orbital images are difficult to maintain a radiometric precision due to the sensor oscillation, atmosphere interferences, season variation of the solar illumination angle, among others. Thus, many radiometric correction techniques have been developed for time series considering mainly: (a) landscape elements whose reflectances are nearly constant over time called of invariants features and (b) linear regression over invariants features assumes that the pixels sampled in the same places at different times are linearly correlated. Therefore, the key problem to the image regression method is an accurate selection of invariant features. In this paper is proposed new radiometric normalization software developed in Turbo C language that searches the highest quality of the invariant features. The algorithm comprises the following steps: (a) identification of the invariant points using a new change detection method based on the spectral classifier algorithms and (b) regression linear between temporal band pairs eliminating the outliers. Initially the algorithm identifies invariants points using a new change detection method based on the spectral classifier algorithms: Spectral Angle Mapper (SAM) and Spectral Correlation Mapper (SCM). In particular, this method approach allows the automatic identification of the invariants points to calibrate remote-sensing images, without visual interpretation data. Program's users establish the spectral change detection method (SAM or SCM) and the threshold value. The second step is to apply two successive linear regressions. First linear regression searches the outlier points using only the pixels with more value than threshold. The outlier points identification use root means square (RMS) and these not include in the second linear regression. Thus, this is last line regression considers only the best spectral for radiometric adjustment. Finally, the gain and offset values are determined and applied for each band in t2 image. In the case of a mistaken selection of points, the program enables to identify a new threshold in both described stages (spectral change detection method and first linear regression).

The present work aims to evaluate the advantages of the ASTER sensor to detect Brazilian Savanna vegetation types over ultramafic or mafic rocks from spectral mixture analysis. The study area, the Niquelandia Complex, lies in the central portion of the State of Goias in Brazil. The Niquelandia Complex is a well-exposed heavily layered intrusion in central Brazil that comprises an area of about 1,800 square km and is approximately 15 km thick. Geology and petrology studies show many similarities between the Niquelandia Complex and well-known Precambrian layered intrusions such as Bushveld and Stillwater. The distribution and physiognomy of vegetation in the Niquelandia Massif show strong geological control. In this site, extensive areas of ultramafic rocks are covered by herbaceous vegetation dominated by grasses. The fires that occur during the driest months (June-August), might influence the physiognomy of the vegetation. However, not only fires can explain the abrupt limit among the gabroic rocks (mafic) and the ultramafic substrate. Different types of forests can be found along the streams and valleys. Their existence demonstrates the ability of many species to tolerate soils originating from ultramafic rock if soil depth and humidity are appropriate and some fire protection exists. The methodology used can be divided into three stages: (a) pre-processing, (b) endmember identification and (c) spectral mixture analysis (SMA). The images were acquired with atmosphere correction relative to the AST07 product. The Level 2 surface reflectance data set (AST07) contains surface reflectance for each of the nine VNIR and SWIR bands at 15-m and 30-m resolutions, respectively. Duplicating the pixels size from the SWIR image the spatial dimensions between VNIR and SWIR images. Endmembers were detected in three steps: (1) spectral reduction by the minimum noise fraction (MNF) transformation, (2) spatial reduction by the Pixel Purity Index (PPI) and (3) manual identification of the endmembers using the N-dimensional visualizer. The spectral classification was made using the Spectral mixture analysis (SMA). The amount of nonphotosynthetic vegetation (NPV) and photosynthetic vegetation (PV) is preponderant in the distinction between the vegetation types. These procedures allowed the identification of the main scenarios in the study area. For the Niquelandia area it was possible to separate for the SMA components vegetation patterns in soils from ultramafic or mafic rocks

Landslides are natural phenomena that model the terrain and cause great damage to humanity, both financially and in terms of loss of life. Therefore predicting this phenomenon is extremely important as it can lead to better city planning and a more effective preventative containment work. Of the various scientific models proposed to date, the SHALSTAB model stands out. It combines a hydrological model and a hillside instability model (based on the Infinite Slope equation). SHALSTAB has been obtaining very good results in areas of the western United States and also in tropical areas. The objective of this work is to define in the terrain, areas more sensitive to landslides and analyze the efficiency of the model using topographical data of different scales. To do this it is necessary to generate a digital elevation model and obtain geomorphologic parameters (such as slope and contribution area) in order to identify the risk areas in the terrain. A high-resolution photogrammetric enlarger was used to create a 1:10,000 scale map of landslide source areas, run off tracks, and deposits. This was then georeferenced to the high resolution DEM. The SHALSTAB performance was verified comparing the unstable pixel in the fall within landslide scat's. The results show that the 80% of the SHALSTAB performance from 1:10,000 scale topographic data was located in unstable classes. However, the model performance from 1:50,000 scale decreases to 50%. In addition, the performance from 1:10,000 scale just 1% of the scar fall at stable class on the other hand around 18% fall in this class when a 1:50,000 scale is used. Thus, we suggest that greater effort and emphasis should be placed on acquisition of high-resolution, high quality topographic. A 50,000 scale can be used in order to determine preliminary landslides hazard areas but it is necessary to have a 1:10,000 scale to increase model performance in critical areas.

The Sao Francisco hydrographic basin comprises seven Brazilian States, representing almost 8% of the country's area, having also different landscapes and huge environmental diversity. The regions of the basin vary form those with high hydrological potentials to those with severe draughts. The irrigation projects on the Sao Francisco River Basin represent an economic growth vector to agricultural activities and agro industries. They would not reach full development under the high climatic risks that are characteristic of the semi-arid climate, which comprises major portions of the basin area. The CODEVASF Brazilian management company has 38 irrigation projects in operation, spread over 31 municipalities of the basin, 10 being implemented and another 10 subjected is viability studies. The objective of this article is to perform a multitemporal analysis, considering the influence of the projects to the economic development of the region. The methodological procedures can be described in three phases: (a) reground development analyses using the Municipal Human Development Index (MHDI) from 1970 to 2000, (b) analysis of the relations between projects implementation and the MHDI. The MHDI is based on factors like income, education and longevity. The MHDI informations were mapped in a GIS. After clustering the municipalities into the MHDI classes, there were introduced maps of the irrigation projects in operation on the basin. It is possible to state that the irrigation projects are not the only factors responsible for an increase in the MHDI in the municipalities along the basin, since all the municipalities showed an increase in the MHDI regardless the presence of the projects. However a significant growth in the MHDI is verified in those municipalities, which had irrigation projects. There is a strong correlation between the growths of the irrigation area with the MHDI. That shows the importance of the projects to the basin as well as the importance of CODEVASF as the head institution for the implementation of such initiatives.

The present work aims to evaluate the advantages of the ASTER sensor for the distinction of Brazilian Savanna vegetation types from spectral mixture analysis. The study area is the Military Instruction Field which has an extensive area with approximately 115.014 ha of native Brazilian Savanna. The methodology used can be divided in three stages: a) preprocessing, b) endmembers identification and c) Spectral Mixture Analysis (SMA). The images were acquired with atmosphere correction. The union of the spatial dimensions between VNIR and SWIR images was done by duplicating the pixels size from the SWIR image. Since the study area is located in two adjacent ASTER images a mosaic was done in order to combine both. Endmembers were detected in three steps: a) spectral reduction by the Minimum Noise Fraction (MNF) transformation, b) spatial reduction by the Pixel Purity Index (PPI) and c) manual identification of the endmembers using the N-dimensional visualizer. The spectral classification was done using the Spectral Mixture Analysis (SMA). The classification was done relative to the different vegetation types and bodies of water with vegetation occurrence. The amount of nonphotosynthetic vegetation (NPV) and photosynthetic vegetation (PV) is preponderant in the distinction between the vegetation types. These procedures allowed identifying the main scenarios in the study area.

This work presents a spectral analysis of natural targets behavior of Cerrado using an AVIRIS sensor image (Airborne Visible/InfraRed Imaging Spectrometer). The sensor AVIRIS was brought to Brazil in 1995 in the SCAR-B (Smoke, Clouds and Radiation-Brazil) mission with the purpose of evaluating atmospheric effects. This activity was accomplished by NASA, INPE (National Institute of Space Research) and AEB (Brazilian Space Agency). During the SCAR-B mission, the AVIRIS sensor acquired image of Goias region, in August 16th 1995. The image used in this study presents geological structures that mark effects on the distribution of vegetation types. The aim of this work was to adapt and to test the employment of the spectral analysis in an AVIRIS hyperspectral image to differentiate vegetation patterns in order to compare with the geological structure. The atmospheric correction was performed using ATREM method (Atmosphere Removal Program) complemented by EFFORT (Empirical Flat Field Optical Reflectance Transformation) method. The endmembers were detected after the atmospheric correction according to the following steps: (a) spectral reduction by the Minimum Noise Fraction (MNF) transformation, (b) spatial reduction by the Pixel Purity Index (PPI) and (c) manual identification of the members using the N-dimensional visualization device. The employment of this technique allowed selecting a spectral series related to vegetation. The spectral analysis showed three main groups: (a) photosynthetic vegetation (PV), (b) non-photosynthetic vegetation (NPV) and (c) burned vegetation. The spectral classification was done using the SCM (Spectral Correlation Mapper) method. The SCM algorithm is a spectral classifier that presents advantages over Spectral Angle Mapper and Spectral Feature Fitting methods due to the ability to detect false positive results. The SCM was performed in the study image using the selected endmember means (photosynthetic active vegetation, NPV and burned areas). In classified images the most similar areas presented lighter colors, while the correlated areas were darkened. The color composition of the SCM classified images allowed the detection of differences in the vegetation distribution, enhancing structural and geomorphologic features of the study site.

Since the 1970s, there is a increasing occupation of the Brazilian "cerrados" by agriculture, with important soil and erosion impacts. The present paper has the purpose to the model morphologic features estimating erosion to subsidize the agricultural planning of the Great river basin (Barreiras, state of Bahia - Brazil), by means of geoprocessing tools. Thus, the topographical factor of Universal Soil Loss Equation (USLE) was used. It was modeled from the Digital Elevation Model (DEM), created by Topogrid module of the ArcInfo software, focusing on hydrology. The slope length of the USLE presents difficulty to be modeled in computational environs. To supply that deficiency methodologies have been proposed in order to obtain automated values with better results. This way, it was developed a program in ArcInfo Macro Language (AML) to optimize the factor slope length map construction from the mathematical formulation proposed before. The input of the developed program it just DEM. A proposition to estimate this variable uses a mathematical formulation from the contribution area. The results maps were: aspect, slope, length exponent of the USLE-factor (m), coefficient X and slope length map. The slope length factor map enables to identify the largest susceptible areas for erosion, improper to agriculture: the results attest that areas prone to erosive process take place at high flow concentration of water.

Landslides are a common problem in the southeast of Brazil mainly after strong summer rainfalls. The understanding of this process has been awaking the interest of scholars. Many methodologies have been developed in order to identify areas which are landslide prone. A methodology named SHALSTAB has been used and/or refined in order to map instability areas on the hillslope. This methodology is a combination between a hydrological model and a slope stability model, based on a digital elevation model (DEM). Recently it was written in Avenue language and implemented for utilization in the ArcView Software. The methodology was applied in the mine region named Quadrila/spl acute/tero Ferri/spl acute/fero in Brazil. A DEM was made and derivatives maps of slope and contributing area were also determined. The soil property parameters were obtained from references. The results demonstrated that the SHALSTAB model is also an effective tool in the identification of susceptible zones for the occurrence of shallow landslides in mine zones.

Growth models of urban clusters are developed to understand the mechanisms of city evolution. Natural, social and economical constraints must be considered in studies of the spatial dynamics of the Latin American cities. However, to describe the urban evolution, methods to generate and handle growth vectors are needed. Great part of this kind of study focuses on qualitative analyses, where trend vectors are manually defined. This research presents an algorithm for the estimation of speed and acceleration of the urban growth. The proposed approach produces results on raster databases, spatially modelled pixel by pixel. This data structure allows integration with other raster base, such as satellite and DEM-derived relief data (slope, altimetry, for example) and others. This method uses the temporal series of the city limits, extracted from satellite imagery. These limits are stored in the vector format labelled with time as z-value, expressed in years, months or days. The interpolation, through kriging techniques, produces a timely based image of the urban growth. Since speed is a relation between ∆s and ∆ ∆ ∆ ∆t (where ∆s is space differential and ∆t the time differential), a growth speed image can be calculated using the expression v=resolution/z, or the inverse of the time image gradient, automatically obtained with the slope angle function. Once acceleration is the relation ∆v/∆ ∆ ∆ ∆t, the same approach can be applied for its calculation. Convolution filtering with 3x3 pixels moving windows on the growth speed image may indicate the higher gradient among the eight windowed neighbours, so as to generate a ∆v image, which can be divided by the ∆t image to produce the acceleration image. Aspect function may be used to generate the directions of the calculated vectors. The developed program automatically executes all these operations to calculate speed and acceleration from the interpolated time image. Tests under varied hypothetical conditions showed the program to perform very well, with promising potentials for the analysis of urban growth, crop expansion, deforestation, gully erosion, as well as any other growth phenomena of different temporal and spatial scales.