Object-based Classification Research Articles

Mapping past landscapes using landsat data: Upper Paraná River Basin in 1985

During the last decades, the science of remote sensing of the Earth's surface has produced an enormous amount of data. In parallel, with the increase in computational capacity, several classification methods have been applied to the satellite retrievals. This timely combination allows recovering more accurate knowledge about the land cover maps of past times. Therefore, the main goal of this work was to develop a land cover product for the year 1985 in the Upper Paraná River Basin (UPRB-1985), one of the largest and most economically important river basins in the world. The land cover map was developed using a supervised classifier - SVM (Support Vector Machine) applied to data from Landsat TM (Thematic Mapper) sensor. The classification process was carried out based on 52 scenes collected during 1985 and a total of 17,040 training samples across the basin. Pixel and Object-based methods were used to classify Landsat scenes. The generated mapping accuracy was assessed using statistical criteria adopted in the literature - Global Accuracy and Kappa Index. The McNemar's test result showed no significant differences (at the 5% level) between the Pixel-based and Object-based classifications, even with the Object-based classification accuracy was slightly higher (Global Accuracy of 79.8%). However, some relationship between the relief and the classification approach was observed. In sub-basins with high slopes, the mean overall accuracy values of the Pixel-based classification approach were 13.1% higher than the Object-based approach. By mapping past land cover, this work is strategic information to understand ongoing processes, as well as to assess changes in land cover that have occurred over time and evaluate to what extent they explain the variability in the hydrology of the region.

Object-Based Wetland Classification Using Multi-Feature Combination of Ultra-High Spatial Resolution Multispectral Images

The Unmanned Aerial Vehicle (UAV) and Google Earth (GE) RGB images have ultra-high spatial resolution. But it is difficult to get a high classification accuracy due to the poor spectral resolution. In this article, the object-based wetland classification is investigated using multi-feature combination of ultra-high spatial resolution multispectral images (MSI). A Gram-Schmidt (GS) transformation is used to fuze Sentinel-2A data with UAV and GE RGB images, respectively, in order to obtain the ultra-high spatial resolution MSI as data sources. Three different feature combination classification scenarios are constructed for fusion GE and UAV MSI, respectively, based on selected features. The object-based random forest (RF) algorithms with parameters (mtry and ntree) optimization are used to carry out finer wetland classification. Results show that the fusion GE and UAV MSI have good applicability in the finer wetland classification, especially the fusion UAV images, and integrating multi-source features could improve classification accuracy. Both data sources reach the highest accuracy in scenario3. The overall accuracy of fusion UAV image scenario3 is 94.31% (Kappa = 0.9353), and that of fusion GE image scenario3 is 87.37% (Kappa = 0.8528). The contribution of different features to wetland classification is obtained with spectral and vegetation indexes, texture, geometric and contextual features.

Monitoring the Mediterranean type forests and land-use/cover changes using appropriate landscape metrics and hybrid classification approach in Eastern Mediterranean of Turkey

Monitoring the Land-Use/Cover Change (LUCC) is an important tool to evaluate the reasons for environmental changes in ecologically sensitive landscapes like natural forestlands. Rural landscapes are of vital importance for ecosystem productivity, ecosystem services, and biological diversity to continue sustainably. The purpose of this paper was to detect LUCC and its effects on landscape ecology through landscape metrics in the Eastern Mediterranean of Turkey. In this study, a hybrid classification approach was used to classify the Land-Use/Cover (LUC) and detailed forest tree diversity considering topography, plant density, and satellite waveband reflectance values. To this extent, detailed LUC classification, LUCC analyses from 2003 to 2014, habitat quality differences by generating landscape metrics in two levels are called landscape and class-level metrics were carried out in the study area. Habitat quality evaluation on forest formation scale using a hybrid classification approach provided a great advantage and made it possible to examine the landscape metrics of the plant types within the scope of temporal change. The study method was implemented in seven stages including: (1) classification of forest–no forestlands with the K-Means algorithm, (2) creating a data set of reflected signals over stand types, (3) determining the rules and thresholds of decision tree algorithm, (4) object-based classification of agricultural, rocky, and settlement areas, (5) obtaining the land-cover maps for 2003 and 2014, (6) post-classification change detection analyses, and (7) assessing the habitat quality via landscape metrics. The results indicated that forest areas increased by 10.73%, while bare soil decreased by 17.70% in 12 years. The habitat quality increased in the same period in the study area according to the results of class area, mean shape index, mean patch size index, edge density, patch number, and Shannon’s diversity index values.

A Comparison of UAV and Satellites Multispectral Imagery in Monitoring Onion Crop. An Application in the ‘Cipolla Rossa di Tropea’ (Italy)

Precision agriculture (PA) is a management strategy that analyzes the spatial and temporal variability of agricultural fields using information and communication technologies with the aim to optimize profitability, sustainability, and protection of agro-ecological services. In the context of PA, this research evaluated the reliability of multispectral (MS) imagery collected at different spatial resolutions by an unmanned aerial vehicle (UAV) and PlanetScope and Sentinel-2 satellite platforms in monitoring onion crops over three different dates. The soil adjusted vegetation index (SAVI) was used for monitoring the vigor of the study field. Next, the vigor maps from the two satellite platforms with those derived from UAV were compared by statistical analysis in order to evaluate the contribution made by each platform for monitoring onion crops. Besides, the two coverage’s classes of the field, bare soil and onions, were spatially identified using geographical object-based image classification (GEOBIA), and their spectral contribution was analyzed comparing the SAVI calculated considering only crop pixels (i.e., SAVI onions) and that calculated considering only bare soil pixels (i.e., SAVI soil) with the SAVI from the three platforms. The results showed that satellite imagery, coherent and correlated with UAV images, could be useful to assess the general conditions of the field while UAV permits to discriminate localized circumscribed areas that the lowest resolution of satellites missed, where there are conditions of inhomogeneity in the field, determined by abiotic or biotic stresses.

Comparison of machine learning algorithms for mapping mango plantations based on Gaofen-1 imagery

Mango is a commercial crop on Hainan Island, China, that is cultivated to develop the tropical rural economy. The development of accurate and up-to-date maps of the spatial distribution of mango plantations is necessary for agricultural monitoring and decision management by the local government. Pixel-based and object-oriented image analysis methods for mapping mango plantations were compared using two machine learning algorithms (support vector machine (SVM) and Random Forest (RF)) based on Chinese high-resolution Gaofen-1 (GF-1) imagery in parts of Hainan Island. To assess the importance of different features on classification accuracy, a combined layer of four original bands, 32 gray-level co-occurrence (GLCM) texture indices, and 10 vegetation indices were used as input features. Then five different sets of variables (5, 10, 20, and 30 input variables and all 46 variables) were classified with the two machine learning algorithms at object-based level. Results of the feature optimization suggested that homogeneity and variance were very important variables for distinguishing mango plantations patches. The object-based classifiers could significantly improve overall accuracy between 2–7% when compared to pixel-based classifiers. When there were 5 and 10 input variables, SVM showed higher classification accuracy than RF, and when the input variables exceeded 20, RF showed better performances. After the accuracy achieved saturation points, there were only slightly classification accuracy improvements along with the numbers of feature increases for both of SVM and RF classifiers. The results indicated that GF-1 imagery can be successfully applied to mango plantation mapping in tropical regions, which would provide a useful framework for accurate tropical agriculture land management.

Hyperspectral Imagery Classification Based on Multiscale Superpixel-Level Constraint Representation

Sparse representation (SR)-based models have been widely applied for hyperspectral image classification. In our previously established constraint representation (CR) model, we exploited the underlying significance of the sparse coefficient and proposed the participation degree (PD) to represent the contribution of the training sample in representing the testing pixel. However, the spatial variants of the original residual error-driven frameworks often suffer the obstacles to optimization due to the strong constraints. In this paper, based on the object-based image classification (OBIC) framework, we firstly propose a spectral–spatial classification method, called superpixel-level constraint representation (SPCR). Firstly, it uses the PD in respect to the sparse coefficient from CR model. Then, transforming the individual PD to a united activity degree (UAD)-driven mechanism via a spatial constraint generated by the superpixel segmentation algorithm. The final classification is determined based on the UAD-driven mechanism. Considering that the SPCR is susceptible to the segmentation scale, an improved multiscale superpixel-level constraint representation (MSPCR) is further proposed through the decision fusion process of SPCR at different scales. The SPCR method is firstly performed at each scale, and the final category of the testing pixel is determined by the maximum number of the predicated labels among the classification results at each scale. Experimental results on four real hyperspectral datasets including a GF-5 satellite data verified the efficiency and practicability of the two proposed methods.

Future scenarios based on a CA-Markov land use and land cover simulation model for a tropical humid basin in the Cerrado/Atlantic forest ecotone of Brazil

Recently, the advancement of agriculture in Brazil has caused very serious problems, such as deforestation and an increase in the amount of pesticides and suspended sediment that have negatively impacted the environment, especially in the Prata River basin, which is located in the Serra da Bodoquena region, Brazil. Thus, the objective of this study was to analyze the future changes in land use and land cover of the advancement of agriculture in the native vegetation areas of the Cerrado/Atlantic forest ecotone in the Prata River basin in 2033, 2050, 2080 and 2100. To map the future land use and land cover (LULC) in the study area, an object-based classification approach was used on Landsat satellite imagery from 1986, 1999, 2007 and 2016. The future LULC scenarios were created with a cellular automata-Markov forecast model. The simulation results of the LULC changes were considered satisfactory, as evidenced by the values obtained from the kappa for agreement (κstandard) = 0.73, kappa for no information (κno) = 0.79 and kappa for grid-cell level location (κlocation) = 0.85 indices. The modeled future scenarios of LULC indicated the advancement of crop agriculture and decreases in wetlands (banhado), savannahs, riparian forests, seasonal semideciduous forests and wet grasslands, which can be considered a warning about the loss of biodiversity of fauna and flora in the Serra da Bodoquena National Park, mainly in the Prata River basin.

Integrating spectral variability and spatial distribution for object-based image analysis using curve matching approaches

Object-based image analysis (OBIA) has been widely used to classify high spatial resolution (HSR) imagery. In a traditional OBIA, object-level statistical summaries such as mean values are usually used for classification. This implies that the spectral values within objects follow a Gaussian distribution. However, the pixel values in an object do not necessarily conform to a Gaussian distribution because of within object spectral heterogeneity. Consequently, these statistical summaries may misrepresent the features of the object. This shortcoming is addressed in this paper by integrating both the spectral variability and the spatial distribution of the pixels within objects to improve the traditional object-based image classification. The spectral variability is represented by histograms of the pixel values in the object, and the spatial distribution is characterized by the binary spatial covariogram of these pixels. To construct a binary spatial covariogram, a principal component analysis (PCA) is first applied to compress multiple bands into one, and the Otsu thresholding is then performed to generate a binary map reflecting the spatial configuration of the pixels. Spatial covariance is then computed for this binary map and plotted with different lag distances to derive the binary spatial covariogram. Our proposed model utilizing curves composed of the spectral histograms and binary spatial covariogram (referred to as the His-Cov model) are then used for classification based on curve matching approaches. The integration of spectral variability and spatial distribution of the pixels in the object produced superior results to curve matching approaches based on spectral variability alone and to traditional OBIA based on spectral and spatial features of the objects when classifying complex land use types in urban environments.

A novel approach for scene classification from remote sensing images using deep learning methods

ABSTRACT Deep learning plays a major task in classification of unsupervised data, which utilises network enhanced learning. This technique proves to be powerful in remote sensing (RS) field for spatial data classification. In the existing environment, huge amounts of data from the earth observation satellites known as satellite images time series (SITS) are gathered, which can be utilised for observing the areas related to geography over through time. In this proposed model the time series model utilised is based on geography. There exists a challenge on how these types of information can be analysed in the field of remote sensing. Notable, techniques related to deep learning substantiated in dealing with remote sensing usually for classification of scene . In this paper, we propose an enhanced classification method involving Recurrent Neural Network (RNN) along with Random forest (RF) for land classification using satellite images, which are publicly available for various research purposes. We utilised spatial data gathered from the satellite images (i.e. time series). Our experimental classification is based on pixel and object-based classification. The attained analysis illustrates that the proposed model outperforms the other present day remote sensing classification techniques by producing 87% target accuracy of classification scene from satellite images.

Forests and urban green areas as tools to address the challenges of sustainability in Latin American urban socio-ecological systems

One sustainability challenge is confronting the process of global urbanization considering wellbeing and environmental justice. Integrating different disciplines to describe and analyze the relationships of urban socio-ecological systems gives us useful empirical tools to formulate policies and programs in urban planning considering their spatial dimension. The objective of this study is to present a model that integrates and explains the socio-ecological urban relations of a Latin American city considering three high-level approaches: forestry, geography, and psychology. Thus, we defined four factors: a) urban forest and green areas; b) urban spatial segregation; c) perceived restoration; and d) subjective wellbeing. For these, we grouped 16 measured variables and collected them with three specific procedures: a) SPOT remote sensors and object-based classification of urban coverage; b) analysis of geospatial data with census information; and c) field surveys. We applied descriptive multivariate statistics and also proposed a structural equation model (SEM) that integrates all the variables and data. We found that the factor “urban green areas” had a direct positive relationship with the factors “urban spatial segregation” and “perceived restoration.” We observed that urban green areas were meeting spaces between different socioeconomic categories, reducing segregation and multiplying opportunities for the psychological restoration of citizens. However, we found no evidence that green areas are related to subjective wellbeing. The model quantified the socio-ecological relationships produced by combining various factors of urban socio-ecological systems, suggesting the benefits of this method to generate knowledge towards planning and managing Latin American cities. Our results are encouraging in terms of environmental justice and wellbeing. In developing countries where forecasts indicate rising urban populations, the need to establish planning processes based on scientific information is vital to meet the challenges of sustainability in the twenty-first century.

A high-resolution remotely sensed benthic habitat map of the Qatari coastal zone

A comprehensive, high resolution, ground truthed benthic habitat map has been completed for Qatar's coastal zone and Halul Island. The objectives of this research were to; 1. Systematically compare and contrast pixel- and object-based classifiers for benthic mapping in a limited focus area and then to, 2. Apply these learnings to develop an accurate high resolution benthic habitat map for the entirety of the Qatari coastal zone. Results indicate object-based methods proved more efficient and accurate when compared to pixel based classifiers. The developed country-wide map covers 4500 km2 and underscores the complex interplay of seagrass, macroalgal, and reefal habitats, as well as areas of expansive mangrove forests and microbial mats. The map developed here is a first of its kind in the region. Many potential applications exist for the datasets collected to provide fundamental information that can be used for ecosystem-based management decision making.

Object-based crop classification in Hetao plain using random forest

Crop classification based on object-based image analysis (OBIA) is increasingly reported. However, it is still challenging to produce high-quality crop type maps by using recent techniques. This article introduces a new object-based crop classification algorithm which contains 4 steps. First, a random forest (RF) classifier is trained by using the initial training set, which tends to have a relatively small size. Second, importance scores for each feature variable are derived by using the RF model. Third, by treating the importance scores as weighting factors, a weighted Euclidean distance criterion is designed and used for sample creation to enlarge training set. Fourth, RF is re-trained by using the enlarged training set, and then it is employed for final classification. To validate the proposed strategy, a Worldview-2 image covering a part of Hetao plain is experimented. Results indicate that the new method yields the best overall accuracy, which equals 90.52%.

Biomass and vegetation coverage survey in the Mu Us sandy land - based on unmanned aerial vehicle RGB images

Accurate detection of vegetation cover and biomass of shrub communities in sandy area is beneficial for evaluating ecosystem, improving remote sensing models, and assessing the accuracy of remote sensing. Unmanned aerial vehicles (UAVs) have replaced traditional measurement methods in biomass and fraction of vegetation coverage (FVC) detection owing to the high spatial resolution of their imagery, their high positioning accuracy, and their ease of use. The existing methods of detecting biomass via UAVs, however, are not suitable for surface fluctuations, dwarf shrubs, and herbs. Futhermore, the method of calculating FVC using UAV RGB images has not yet been tested in sandy areas. To accurately extract FVC data, aboveground biomass (ABS) and net aboveground biomass (NABS) of shrub communities in the desert regions, UAV RGB images of 87 sample plots in the Mu Us sandy land were collected and used to obtain the FVC and biomass information via the object-based classification method, single shrub canopy biomass model and vegetation index-based method. The results are as follows: (1) the method of calculating ABS and NABS based on shrub canopy width extraction can be used in desert shrub communities, and results show that the ABS and NABS of vegetation communities increases from 15 to 800 g/m2 and 10–250 g/m2, respectively, in the Mu Us sandy land; and (2) the lowest value of ABS (NABS) appeared in the mobile sandy dunes and the highest value appeared in the semi-fixed sandy dunes; (3) under fixed thresholds conditions, the FVC can be extracted accurately using the excess green method, visible atmospherically resistant index, vegetative index, green red vegetation index and red green blue vegetation index (RGBVI); and (4) the correlation between the FVC calculated by the five RGB vegetation indexes and NABS in this study is greater than that between FVC and ABS (e.g. R2NABS - RGBVI = 0.734, R2ABS - RGBVI = 0.666), and the FVC calculated by RGBVI can be used to estimate NABS in the Mu Us sandy land. This study will provide new insights for field investigations of the ABS, NABS, and FVC in sandy areas.

Mapping Tea Plantations from VHR Images Using OBIA and Convolutional Neural Networks

Tea is an important economic plant, which is widely cultivated in many countries, particularly in China. Accurately mapping tea plantations is crucial in the operations, management, and supervision of the growth and development of the tea industry. We propose an object-based convolutional neural network (CNN) to extract tea plantations from very high resolution remote sensing images. Image segmentation was performed to obtain image objects, while a fine-tuned CNN model was used to extract deep image features. We conducted feature selection based on the Gini index to reduce the dimensionality of deep features, and the selected features were then used for classifying tea objects via a random forest. The proposed method was first applied to Google Earth images and then transferred to GF-2 satellite images. We compared the proposed classification with existing methods: Object-based classification using random forest, Mask R-CNN, and object-based CNN without fine-tuning. The results show the proposed method achieved a higher classification accuracy than other methods and produced smaller over- and under-classification geometric errors than Mask R-CNN in terms of shape integrity and boundary consistency. The proposed approach, trained using Google Earth images, achieved comparable results when transferring to the classification of tea objects from GF-2 images. We conclude that the proposed method is effective for mapping tea plantations using very high-resolution remote sensing images even with limited training samples and has huge potential for mapping tea plantations in large areas.

Automatic shoreline detection and its forecast: a case study on Dr. Abdul Kalam Island in the section of Bay of Bengal

For 43 years, multi-dated satellite imagery is used to examine shifts in shorelines and potential location on shorelines for Dr. Abdul Kalam Island, in the Bay of Bengal, India. A Normalized Difference Water Index with object-based classification system and automatic shoreline extraction tool is used to retrieve the shoreline automatically. To calculate the rates of shoreline change and future positions, based on empirical observations, the simple statistical model, such as EPR and LR models are used. The erosion/accretion scenario is also analyzed for the 1976, 1989, 1997, 2008 and 2018 shoreline of Landsat imaging.Results showsnorth of Dr. Abdul Kalam Island is subjected to high deposition rates and the south part having high erosion rates in the vicinity of the Dhamara Estuary and Maipura River. The short-term (2023) and long-term (2028) shoreline positions are predicted on the basis of the delineated shoreline. Around 63% of transect’s RMSE values varies between ±10 m, showing better agreement among the estimated shoreline positions and satellite-based ones. The results of the study can be very useful for the quantification of shoreline changes, shoreline forecasting and precautionary measures for anti-satellite missile launching centre.

Assessing the effect of water scarcity on crop selection and spatial pattern of croplands in central Iran

Farming and food production for the increasing world’s population require proper preparation of a diverse range of environmental parameters especially water availability and accessibility. This study investigated the effect of water scarcity on crop selection and spatial pattern of croplands in an agricultural landscape in central Iran. Towards this aim, the main crops including wheat, alfalfa, potato, fruit trees, rice and vegetables were mapped using a combination of decision tree and object-based classifiers and four Landsat 8 images acquired from March 2015 to June 2015 with Kappa Coefficient and Overall Accuracy of 80.16 and 85.37 percent, respectively. Groundwater depth as an indicator of water availability was mapped by interpolating over 13 thousand wells and, according to the ability of farmers and available technologies to extract water, classified into four groundwater level areas (GLA). Finally, spatial pattern of crops in each GLA was investigated using four landscape-level metrics including Mean Patch Size (MPS), Mean Nearest-Neighbor distance (MNN), Contagion index (CONTAG) and Shannon's diversity (SHDI). The results of this study showed that cropping system in drier areas has shifted towards monoculture (crops with fewer water requirements such as wheat) and led to a fragmented and homogeneous landscape with large and low dispersed agricultural units. It was finally concluded that changes in crop selection and spatial pattern of croplands could be considered and applied as a precise and fast indicator of determining water availability and accessibility over large geographical regions in Iran.

Spatiotemporal analysis of urban growth on forest and agricultural land using geospatial techniques and Shannon entropy method in the satellite town of Ethiopia, the western fringe of Addis Ababa city

Rapid conversion of conserved land and cropland to non-agricultural purposes is threatening the ecological areas and dominant agricultural activities that are the main sources of livelihood in urban fringe areas of Addis Ababa City. The combinations of government policy, socioeconomic, demographic dynamics, and biophysical triggers have driven such transformation. However, the recent fast urban expansion and infrastructural development could have accelerated the severity and rate at which urban growth impacts the ecosystem and fertile agricultural land. Yet, little attention is given about their impacts on forest and farming communities in the western fringe areas of Addis Ababa during the recent past. This study, therefore, aimed at quantifying and analyzing the trends of the urban growth and its impacts on flora and agricultural land in Sebeta-Awas town using an integrated GIS tool, remote sensing technique, and Shannon entropy method. Landsat TM of 1986, ETM+ of 2002, and OLI of 2019 were used to produce land use/land cover (LULC) classes. Object-based classification technique was carried out to generate the LULC and to measure the changes in the urban land-use class within the satellite town in the year 1986, 2002, and 2019. Shannon entropy method was applied to model study area’s urban sprawl, growth trend, and spatial change. Over the past three decades (1986–2019), Sebeta-Awas town has experienced severe urban sprawl following lack of proper development control and management. The annual urban growth rates of 1.2, 5.5, and ~15% for the periods of 1986, 2002, and 2019, respectively were obtained mainly at the expenses of agricultural land (25.48%) and forest land (16.6%), catalyzed urban sprawl which finally led to serious deforestation and reduction in rural farmland. The findings indicate that the average of entropy index increased from 0.02 in the year 1986 to 0.996 in the year 2019, indicating more dispersed urban growth to the outskirts, and spatially indicating anticlockwise shifting. In this regard, more than 90.2% of forest loss due to agricultural encroachment, built-up area expansion, and construction was widely observed in Sebeta and Alem Gena areas, of course, Northeast zone in this study. It is concluded that deforestation and continual evacuation of farming communities in the urban fringe areas due to rapid urban expansion in the name of investment and infrastructural development is expected to be worsened in the near future unless strong policy revision and management actions are undertaken.

UNSUPERVISED SEGMENTATION OF INDOOR 3D POINT CLOUD: APPLICATION TO OBJECT-BASED CLASSIFICATION

Abstract. Point cloud data of indoor scenes is primarily composed of planar-dominant elements. Automatic shape segmentation is thus valuable to avoid labour intensive labelling. This paper provides a fully unsupervised region growing segmentation approach for efficient clustering of massive 3D point clouds. Our contribution targets a low-level grouping beneficial to object-based classification. We argue that the use of relevant segments for object-based classification has the potential to perform better in terms of recognition accuracy, computing time and lowers the manual labelling time needed. However, fully unsupervised approaches are rare due to a lack of proper generalisation of user-defined parameters. We propose a self-learning heuristic process to define optimal parameters, and we validate our method on a large and richly annotated dataset (S3DIS) yielding 88.1% average F1-score for object-based classification. It permits to automatically segment indoor point clouds with no prior knowledge at commercially viable performance and is the foundation for efficient indoor 3D modelling in cluttered point clouds.

Flood Mapping Using Relevance Vector Machine and SAR Data: A Case Study from Aqqala, Iran

The use of satellite imagery to monitor flood areas is essential to determine the damage and prevent related problems in the future. This paper examines thresholding and unsupervised classification for flood mapping using Sentinel-1 SAR image. Thresholding helps us to determine over-detection and under-detection regions in the flooded areas, and so, gamma distribution is used to select the thresholds. Also, the relevance vector machine (RVM) and the object-based classification method have been used for classification. The RVM algorithm obtained better results with overall accuracy = 0.89 and k = 0.95, while for the object-based classification method, these values were 0.87 and 0.91, respectively. According to the results, over- and under-detection occurred in flat areas and man-made structures, respectively. The results demonstrate a great potential of radar imagery for operational detection and delimitation of water in flood risk areas. The automation of satellite radar data processing operation has been tested, and it shows a potential for optimising the system of monitoring and early detection of flood risk.

BUILDING EXTRACTION FROM VERY HIGH-RESOLUTION SATELLITE IMAGES FOR MAP UPDATING IN EGYPT

Robust building detection from satellite images has been a subject of interest for several decades. Very High Resolution (VHR) satellite images support the efficient extraction of manmade objects. The main aim of this paper is to present an approach for building extraction from VHR satellite images for map updating in Egypt. To achieve this aim, a comparison of pixel and object-based classification techniques has been applied. Then, different refinement processes based on shadow, context, shape, and Digital Surface Model (DSM) data are carried out. Two study areas from the VHR satellite images for Assuit and Sohag cities are used. A comparison of the classification techniques shows that the Maximum Likelihood Classifier (MLC) for pixel-based technique and Support Vector Machine (SVM) for object-based technique give the highest overall accuracy results. Refinement based on shadow, context, shape, and DSM information improves the overall accuracy with an average of 18%. Thus, the building extraction results can contribute significantly to update maps in Egypt.