Recent advances in deep learning architectures, driven by Artificial Intelligence (AI), have significantly enhanced time-series forecasting accuracy. However, the inherent lack of interpretability and causal reasoning in black-box models limits their adoption in high-impact applications such as traffic management and urban mobility. To address this challenge, we propose$\mathrm{C}^{2}$-STORM (Causal Counterfactual Spatio-Temporal TransfORmer Model), a novel framework that integrates deep learning to achieve both accurate and interpretable predictions. Our approach uses frontdoor-adjusted causal extraction to reduce confounding bias, grounded in structural causal principles. This is coupled with a Dynamic Causality Generator that learns context-aware Spatio-Temporal (ST) dependencies via graphstructured attention, enhancing node-specific feature prioritization. Furthermore, we develop a transformer-based counterfactual reasoning module that simulates actionable “what-if” scenarios, aligning interventions with domain-specific logic. Experiments conducted on real-world datasets (PeMSD7, LondonBike) demonstrate the effectiveness of$\mathrm{C}^{2}$-STORM, achieving state-of-the-art performance with improvements of 4.2- 6.6% in MAE/MAPE over baseline methods. Ablation studies confirm the significance of each component, while visual explanations, including causal graphs and feature importance maps, enhance the interpretability and transparency of predictions. By bridging causal inference with ST learning, C²-STORM represents a significant advancement in trustworthy forecasting for complex systems.

Effectively characterize the current land cover status is crucial for assessing agricultural production, monitoring natural resources, and making informed land management decisions. Satellite Image Time Series (SITS) data, capturing spatio-temporal information, are the primary source of information to support the general task of Land Use Land Cover (LULC) mapping. With the aim to effectively exploit the spatio-temporal information carried out by SITS data for the underlying task of land cover mapping, here we introduce a multi-scale classification framework that combines together pixel-level and object-level multivariate SITS information with the objective to ameliorate the pixel-level time series analysis. To assess the behaviour of the proposed framework, we provide a comparative analysis with several SITS-based land cover mapping strategies on a challenging study area, namely Koumbia, located in the Burkina Faso. The obtained results reveal that the joint use of pixel-level and object-level information clearly ameliorates the classification performances.

This study aims to assess the severity of desertification in Tunisia by using machine learning and deep learning models to analyze remote sensing data. By employing MODIS ecological indicators such as Modified Soil-Adjusted Vegetation Index (MSAVI), Normalized Difference Vegetation Index (NDVI), Albedo and Topsoil Grain Size Index (TGSI), different models including eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Decision Tree (DT), LGBM, Convolutional Neural Networks (CNN), Variational Autoencoder (VAE) and Recurrent Neural Networks (RNN) will be trained and compared to effectively monitor desertification trends in Tunisia from 2016 to 2022. The evaluation using metrics such as accuracy, F1 score, recall, and precision showed that XG-Boost performed best with an accuracy of 92.35%. The study also identified a negative trend of desertification affecting the entire country. This research highlights the importance of using computer vision to monitor and address the challenges of desertification and suggests potential applications for environmental management.

The assessment of satellite products over agricultural areas has several challenges that get more complex in developing countries, particularly when exploiting active observation. The use of free-open access satellite images such as Sentinel-1 (S-1) and Sentinel-2 (S-2) images as data sources is considered a great alternative to minimize the high cost and time consumption of the field measurements. The composition of the rainfed agricultural zones depends on the vegetation and soil layers, in that sense, the ability to detect changes in surface parameters is related to the adequate representation of these components. The selection of vegetation descriptors to represent the vegetation changes is one of the main steps for accurate simulations over vegetated areas. Therefore, it is essential to assess the impact of the different vegetation indices (VIs) derived from optical information when included in active models. In this study, the Normalized Difference Vegetation Index (NDVI) and Specific Leaf Area Vegetation Index (SLAVI) were compared with the Vegetation Water Content calculated based on NDVI values (VWC-ndvi). In this paper, the Water Cloud Model (WCM) was used based on in-situ information collected during a field experiment conducted in a rain-fed agricultural area located in Central Mexico (THEXMEX-19). It was concluded that the use of the VW C-ndvi obtained 0.97, -0.11 dB, and 0.49 dB for correlation coefficient (r), Bias, and root mean squared difference (RMSD), respectively. Based on the results, this work highly recommends the use of the WCM and Oh model when evaluating different combinations of parameters for operational use, especially for agricultural purposes.

Deep Learning (DL) has demonstrated remarkable success in various Remote Sensing Image (RSI) analysis applications. However, due to disparities in data distributions, DL models find it challenging to generalize meaningfully, especially when training and testing datasets are collected at different locations with varying resolutions, by different sensors, or due to climatic conditions. DL techniques applied to RSI have shown interest in domain adaptation as a suitable solution for addressing discordance among domains. In this study, we focus specifically on two DL approaches for Domain Adaptation (DA) in RSI: Self-Supervised Learning (SSL) and Graph Neural Networks (GNNs). First, we elucidate the motivation for utilizing DA techniques to address challenges in the field of RSI, along with their applications in conjunction with GNNs and SSL. Then, we present related surveys on domain adaptation and provide background information. This paper suggests a classification system for DL approaches and draws attention to challenges and research directions for DA in RSI. This study aims to deliver scholars in the remote sensing field with current references on DA using SSL and GNNs.

Accurate forecasting of climate and vegetation changes through SDG 13 initiatives is critical for adapting agricultural practices, as it allows farmers to anticipate changes in growing conditions and vegetation patterns, ensuring long-term food security. In this paper, a data science hybrid pipeline is proposed to analyse and forecast the impact of climate change on vegetation using remote sensing data. The proposed pipeline follows a three-step process. In the first step, different data (i.e. normalized difference vegetation index (NDVI), standardized precipitation index and land cover) are collected and pre-processed. Subsequently, the second step involves feature extraction by analysing NDVI and SPI data based on the wavelet transform method and analysing the change detection using land cover information. Finally, the third step completes the process by exploring the extracted feature based on the wavelet transform to forecast the vegetation change based on the temporal convolution network (TCN) model. This pipeline performs and validates various analyses related to vegetation health, including trend component extraction, the timing of maximum vegetation index, response of vegetation changes to meteorological variables and forecasting vegetation change based on climatic data. These results, obtained for Tunisia, Turkey, and Spain, will empower decision-makers to improve agricultural practices.

Soil moisture is a critical parameter that is relevant to several activities such as agriculture and disaster management (e.g., drought, floods, water stress periods, etc.) [1]. Thus, it has a considerable impact on life on Earth. Recently, researchers are counting on the advanced development of remote sensing (RS) technologies for soil moisture retrieval. RS techniques provided amounts of free high, spatial, and time series images that serve soil moisture retrievals at the agricultural field scale such as Synthetic Aperture Radar (SAR). Therefore, this work aims at evaluating the impact of different formulations to represent the soil contribution using synergistically both Sentinel-1 and Sentinel-2 images. This evaluation will reveal the impact of the different soil formulations to represent the backscatter from the ground within the Water Cloud Model (WCM). The formulations representing the soil contribution in this work are the linear equation and the Oh model. The evaluation experiments are set based on in-situ data collected from different fields in Huamantla, Central Mexico. The statistical evaluation of the different formulations of the soil contribution in the total backscatter from the WCM is obtained by comparing the simulations with satellite observations over the complete period. The best results of the evaluated formulations are recorded by the Oh model in the Alvaro site with bias and root mean squared difference (RMSD) values equal -0.0002 dB and $ - {1.49_{x{{10}^{ - 6}}}}$ dB for VV polarization and, 0.0011dB, and ${8.1452_{x{{10}^{ - 6}}}}$ dB for VH polarization. In contrast, the best results obtained by the linear equation are obtained in the Macario site with bias and RMSD values equal to 0.6622 dB and -0.724dB, for VV polarization and 1.2990 dB, and 1.2994 dB for VH polarizations. On the other hand, the total backscatter from the WCM combined with the Oh model outlined the highest accuracy during the vegetated period. In the Palafox-2 site, the combined model obtained a correlation coefficient (r), bias, and RMSD up to 0.965 dB and, -0.107 dB and 0.494 dB for VV polarization, respectively, and 0.916 dB, 0.572 dB, and 1.655 dB for VH polarization, respectively. The slight differences between the two surface scattering models within the WCM suggest that both formulations could be suitable to be implemented in a retrieval process, depending upon the available ancillary information.

This paper proposes a method for automatically monitoring and analyzing the evolution of complex geographic objects. The objects are modeled as a spatiotemporal graph, which separates filiation relations, spatial relations, and spatiotemporal relations, and is analyzed by detecting frequent sub-graphs using constraint satisfaction problems (CSP). The process is divided into four steps: first, the identification of complex objects in each satellite image; second, the construction of a spatiotemporal graph to model the spatiotemporal changes of the complex objects; third, the creation of sub-graphs to be detected in the base spatiotemporal graph; and fourth, the analysis of the spatiotemporal graph by detecting the sub-graphs and solving a constraint network to determine relevant sub-graphs. The final step is further broken down into two sub-steps: (i) the modeling of the constraint network with defined variables and constraints, and (ii) the solving of the constraint network to find relevant sub-graphs in the spatiotemporal graph. Experiments were conducted using real-world satellite images representing several cities in Saudi Arabia, and the results demonstrate the effectiveness of the proposed approach.

The Red Palm Weevil (RPW) is a highly destructive insect causing economic losses and impacting palm tree farming worldwide. This paper proposes an innovative approach for sustainable palm tree farming by utilizing advanced technologies for early detection and management of RPW. Our approach combines computer vision, deep learning (DL), the Internet of Things (IoT), and geospatial data to effectively detect and classify RPW-infested palm trees. The main phases include; (1) DL Classification using sound data from IoT devices, (2) palm tree detection using YOLOv8 on UAV images, and (3) RPW mapping using geospatial data. Our custom DL model achieves 100% precision and recall in detecting and localizing infested palm trees. The integration of geospatial data enables the creation of a comprehensive RPW distribution map for Efficient monitoring and targeted management strategies. This technology-driven approach benefits agricultural authorities, farmers, and researchers in managing RPW infestations, safeguarding palm tree plantations’ productivity.

The United Nations (UN) 2030 Agenda involves 17 major Sustainable Development Goals (SDGs). These SDGs have great implications for country-wide development and making plans in both developed and developing nations in the post-2015 period to 2030. The SDGs are a set of 17 interlinked global goals designed to be a plan to attain a better sustainable future by the combination of earth observation (EO) and artificial intelligence architecture. To attain the goal of global sustainable protection and utilization of terrestrial ecosystems, it is important to quantitatively determine the implementation of Sustainable development goal 15 (SDG-15) and goal 13 (SDG-13). In this paper, we focus on the integration of these SDGs in Tunisia as a new regional development plan. Thus, we present a complete benchmark that aims to solve the complicated analytical problems related to the sophisticated data type using Deep learning (DL) architecture.

Forecasting vegetation indices represents a hot topic nowadays because of the climate change impacts challenges. Several time series forecasting models are proposed in the literature. However, researchers are continuously looking for an accurate and more performing one. This paper proposes a novel way to combine forecasting models results. Particularly interested in vegetation changes, our study relies on NDVI time series extracted from MODIS DATA. These temporal profiles describe a forest cover located in the northwestern of Tunisia, an African country. To this aim, normally fitted weights and evidence theory were selected to build an accurate combination model. Our findings highlight: on one hand, the importance of studying the statistical properties of the analyzed data. On the other hand, they shed the light on the ability of increasing the accuracy of forecasting models by combining them.

Desertification detection is a challenging problem because of the dynamic climate change and human activity. This paper proposes a methodology to detect regions with desertification risk based on Landsat imagery (RGB bands and NDVI) and variational auto-Encoder (VAE). The considered features (RGB and NDVI) are extracted from multitemporal Landsat optical images taken from the freely available Landsat program from 2001 to 2020. The arid region around Jeffara in Medenine (Tunisia) is selected as a study area. The VAE model was evaluated and compared with two deep learning models: convolutional neural network (CNN), convolutional recurrent neural network (CNN_RNN), and VAE without NDVI. The comparative results showed that the VAE outperformed the other models for desertification detection, with an accuracy of over 98 %.

Over the past few years, total financial investment in the agricultural sector has increased substantially. Palm tree is important for many countries’ economies, particularly in northern Africa and the Middle East. Monitoring in terms of detection and counting palm trees provides useful information for a variety of stakeholders; it helps in yield estimation and examination to ensure better crop quality and prevent pests, diseases, better irrigation and other potential threats. Despite their importance, these information still difficult to obtain. In this study, we systematically review research articles between 2011 and 2021 on artificial intelligence (AI) technology for smart palm tree detection. A systematic review (SR) was performed using the PRISMA approach based on a four-stage selection process. Twenty-two articles were included for the synthesis activity reached from the search strategy alongside the inclusion criteria in order to answer tow two main research questions. The study's findings reveal patterns, relationships, networks, and trends in the application of artificial intelligence in the palm tree detection over the last decade. Overall, despite the good results achieved in most of the studies, the effective and efficient management of large-scale palm plantations still a challenge. In addition, countries which their economy strongly related to intelligent palm services especially in North Africa should give more attention to this kind of studies. The results of this research could benefit both the research community and stakeholders.

Soil moisture (SM) is an important parameter used to control a broad range of environmental applications. An increasing attention has been recently given to machine learning (ML) methods for SM retrieval that provide promising performance. Nevertheless, most of them are based on a supervised learning strategies that depend on the used labeled training samples. In fact, they are unaffordable or costly. In this letter, new teacher–student for SM estimation, called (TS-SME), relying on teacher–student (TS) framework using synthetic aperture radar (SAR) and optical data, was proposed to estimate SM. The main advantage of this framework is to enroll a large amount of unlabeled data together with a small amount of labeled data. Experiments were carried out on two arid areas in southern Tunisia. The input data include the backscatter coefficient in two-mode polarization ($\sigma ^{\circ }_{\textrm {VV}}$and$\sigma ^{\circ }_{\textrm {VH}}$) for Sentinel-1A, normalized difference vegetation index (NDVI) and normalized difference infrared index (NDII) for Sentinel-2A andin situmeasurements. Extensive experimental results demonstrated that TS-SME framework is capable of generating a well-performed student model, with the estimation accuracy is superior to all teacher models [artificial neural network (ANN), eXtreme gradient boosting (XGBoost), random forest regressor (RFR), and water cloud model (WCM)]. It was highly correlated with thein situmeasurements with high Pearson’s correlation coefficient$R$(${R}_{\textrm {RF}} =0.86$,${R}_{\textrm {ANN}} =0.75$,${R}_{\textrm {XGBoost}} =0.77$,${R}_{\textrm {WCM}} =0.77$,${R}_{{\,\,\textrm {TS-SME}}} =0.96$) and low root mean square error (RMSE) ($\textrm {RMSE}_{\textrm {RF}} =1.09$%,$\textrm {RMSE}_{\textrm {ANN}} =1.49$%,$\textrm {RMSE}_{\textrm {XGBoost}} =1.39$%,$\textrm {RMSE}_{\textrm {WCM}} =1.12$%,$\textrm {RMSE}_{\,\,\textrm {TS-SME{} }} =0.8$%), respectively.

Soil Moisture (SM) is an important parameter for many environmental applications. In recent years, different Machine Learning (ML) methods have been developed in order to estimate the SM. The aim of this paper is to evaluate three ML methods based methodology to estimate the SM, namely Artificial Neural Networks (ANN), Random Forest Regression (RFR) and extreme Gradient Boosting (XGBoost). The used input data for the three ML methods include the in situ measurements SM, the Backscatter coefficient in two modes polarizations (σ°yy and σ°YH) from Sentinel-1A and Normalized Difference Vegetation Index (NDVI) from Sentinel-2A data. The comparative results show that the SM obtained using the ANN, RF and XG Boost algorithms are highly correlated to the in-situ SM, with high Pearson's correlation coefficient R (RRF= 0.86, RANN= 0.75, R XGBoost= 0.77) respectively and low RMSE (RMSERF= 1.09%, RMSEANN= 1.49%, RMSExGBoost= 1.39%) respectively.

Constraint satisfaction problem (CSP) has been actively used for modeling and solving a wide range of complex real-world problems. However, it has been proven that developing efficient methods for solving CSP, especially for large problems, is very difficult and challenging. Existing complete methods for problem-solving are in most cases unsuitable. Therefore, proposing hybrid CSP-based methods for problem-solving has been of increasing interest in the last decades. This paper aims at proposing a novel approach that combines incomplete and complete CSP methods for problem-solving. The proposed approach takes advantage of the group search algorithm (GSO) and the constraint propagation (CP) methods to solve problems related to the remote sensing field. To the best of our knowledge, this paper represents the first study that proposes a hybridization between an improved version of GSO and CP in the resolution of complex constraint-based problems. Experiments have been conducted for the resolution of object recognition problems in satellite images. Results show good performances in terms of convergence and running time of the proposed CSP-based method compared to existing state-of-the-art methods.

The amount of remote sensing (RS) data has increased at an unexpected scale, due to the rapid progress of earth-observation and the growth of satellite RS and sensor technologies. Traditional relational databases attend their limit to meet the needs of high-resolution and large-scale RS Big Data management. As a result, massive RS data management is currently one of the most imperative topics. To address this problem, this paper describes a distributed architecture for big RS data storage based on a unified metadata file, pyramid model, and Hilbert curve for data composition and indexing using NoSQL databases (i.e, Apache Hbase). In this paper, a Hadoop-based framework in AzureInsight cloud platform is designed to manage massive RS data in a parallel and distributed way. Experimental results prove that our method has the potential to overcome the weakness of traditional methods. The proposed model is suitable for massive high-resolution image data management.

Sub-pixel mapping (SPM) denotes a category of image processing techniques that further enhance the results provided by spectral unmixing algorithms. While the latter is only capable to determine the fractional abundances of classes with an associated spectral signature within a certain area denoted as mixed pixel, SPM can in addition spatially locate each class separately within the mixed pixel itself, enhancing the spatial resolution of its products. Given the demands by both technological and scientific application, various approaches have been proposed; this work will focus on the ones belonging to the domain of variational frameworks, which allow to solve the intrinsic ill-posedness of inverse problems by imposing a regularization based on gradients of the desired output. As the problem of SPM may be also seen as generating a rule to associate each mixed pixel to a specific patch of mosaicked classes, we propose to create an overcomplete dictionary listing them. We present here some first investigation in this direction, by incorporating such dictionary, generated via K-SVD dictionary learning algorithm, in the formulation of an inverse problem. The algorithm is matched with a Isotropic Total Variation (ITV) regularization to provide joint spatial and spectral consistency to the results. Our tests prove that our proposed method provides better SPM product quality compared to its use in insulation and to other state of the art variational framework-based algorithms.

Hyperspectral imaging is an important source of land cover information by virtue of its spectral richness. However, this type of imagery is typically known by its coarse spatial resolution, that is a limiting factor for end-users. Although spectral unmixing techniques can provide subpixellic information by means of abundance fractions for each class in mixed pixels, the spatial distribution of these classes within each pixel is still unknown. Sub-pixel mapping techniques address the above mentioned problem. Nevertheless, the traditional sub-pixel mapping algorithms based on spatial dependence assumptions cannot solve these problems efficiently. Spatial regularization methods have recently been proposed in a way that they can treat each abundances map separately and do not consider spatial correlation between classes. In order to improve sub-pixel mapping accuracy and, consequently, enhance hyperspectral image classification, we propose a sub-pixel mapping method based on isotropic total variation minimization within and between pixels for different classes simultaneously. Experimental results with synthetic data sets show the attributes of using total variation as a prior model, which leads to improve sub-pixel mapping of different classes together.

3D urban models provide convincing analytic tools for decision making, city planning, and smart city services. However, developing a fully automated method that can produce 3D building models of high quality, fidelity and accuracy is still a challenging task. Currently, most of the proposed approaches handle polyhedral roofs (consisting of planar polygons) because they assume that all roofs in a single area follow this prior. However, the reconstruction method could have its prior adapted to the roof type. In this paper, we are dealing with a specific roof case which is piecewise horizontal roofs which are very frequent in most countries of North Africa and in particular in Tunisia. Our building reconstruction method follows four main steps: building LiDAR points extraction, piecewise horizontal roof clustering, boundary creation and 3D geometric modeling. In order to prove the suitability and the effectiveness of the introduced method, experiments are conducted with real LiDAR data and aerial RGB image.

This research deals with semantic interpretation of Remote Sensing Images (RSIs) using ontologies which are considered as one of the main challenging methods for modeling high-level knowledge, and reducing the semantic gap between low-level features and high-level semantics of an image. In this paper, we propose a new ontology which allows the annotation as well as the interpretation of RSI with respect to natural risks, while taking into account uncertainty of data, object dynamics in natural scenes, and specificities of sensors. In addition, using this ontology, we propose a methodology to (i) annotate the semantic content of RSI, and (ii) deduce the susceptibility of the land cover to natural phenomena such as erosion, floods, and fires, using case-based reasoning supported by the ontology. This work is tested using LANDSAT and SPOT images of the region of Kef which is situated in the north-west of Tunisia. Results are rather promising.

High-resolution remote sensing scene classification (HR-RSSC) plays an increasingly important role since it aims to enhance the scene semantic understanding. Recently, convolutional neural networks (CNNs) proved their effectiveness in learning powerful feature representations for various visual recognition tasks. However, in the RS domain, the performance of CNN is still limited due to the lack of sufficient labeled data. In this letter, we propose an HR-RSSC method based on CNN transfer learning (TL) for feature extraction (FE) and deep forest (DF) for classification. In fact, we extract deep features from the last convolutional layer in order to avoid the use of the fully connected layers (FCLs) which need many parameters to tune. Moreover, we train a DF model that is based on ensemble learning that can achieve better performances than single classifiers and is easy to train with few parameters. We evaluate the proposed method on two RS image. Compared to full-training, fine-tuning, and state-of-the-art CNN TL methods, the results demonstrate the effectiveness of the DF model for HR-RSSC based on CNN TL in terms of overall accuracy and training time.

Remotely sensed images represent an important source of information for monitoring land changes that may occur. There is, therefore, a need to analyze and interpret such information in order to extract useful semantic change interpretations. However, extracting such semantics from satellite images is a complex task that requires prior and contextual knowledge. In this paper, we focus on the issue of semantic scene interpretation for change interpretation. Consequently, a strategy for semantic remote-sensing imagery scene interpretation is proposed. This strategy is based on a representative framework that is structured around several levels of interpretation: the pixel level, the visual primitive level, the object level, the scene level, and the change interpretation level. Each level integrates a logical mechanism to extract useful knowledge for interpretation. The proposed model has been evaluated using two Landsat scene images acquired in 2000 [Landsat Enhanced Thematic Mapper plus (ETM+)] and 2017 (Landsat 8) in order to check its relevance for semantic scene and change interpretation. Precision, recall, and F-measure metrics were used in order to show the capacity of the proposed methodology for semantic classification. A visual evaluation was also performed to evaluate the performance of the presented interpretation strategy, and the query results for each level show a promising capability for semantic object classification, spatial and temporal relations' extraction, and change interpretation.

This paper addresses the problem of human action recognition from sequences of 3D skeleton data. For this purpose, we combine a deep learning network with geometric features extracted from data lie on a non-Euclidean space, which have been recently shown to be very effective to capture the geometric structure of the human pose. In particular, our approach claims to incorporate the intrinsic nature of the data characterized by Lie Group into deep neural networks and to learn more adequate geometric features for 3D action recognition problem. First, geometric features are extracted from 3D joints of skeleton sequences using the Lie group representation. Then, the network model is built from stacked units of 1-dimensional CNN across the temporal domain. Finally, CNN-features are then used to train an LSTM layer to model dependencies in the temporal domain, and to perform the action recognition. The experimental evaluation is performed on three public datasets containing various challenges: UT-Kinect, Florence 3D-Action and MSR-Action 3D. Results reveal that our approach achieves most of the state-of-the-art performance.

In this paper we tackle the problem of data analysis on high dimensional space of hyperspectral images. The remote sensing data analysis is a complex task due to many factors such as the large spectral and spatial variability. In fact, several approaches suffer from the pixel-mixed problem since mixed pixels are often sources of uncertainty and inaccuracy. Although spectral un-mixing techniques can provide abundance fractions for each class in mixed pixels, spatial distribution of these classes remains still unknown. Sub-pixel mapping techniques provide any solutions to the above-mentioned problem. Nevertheless, most of the proposed methods treat one particular type of objects because they assume that mixed pixels in a single image are identical. Sub-pixel mapping methods employing zonal objects are based on the spatial dependence assumption; the same assumption is used for linear objects while considering objects direction. Whereas, in case of encapsulated objects no spatial correlation is applied. As a result, each type of object requires special treatment, hence, the need for object recognition at sub-pixel scales. In this paper to improve sub-pixel object recognition accuracy in coarse spatial resolution image, we develop in this paper a spectro-spatial method based on a graph model, which discriminates between each type of sub-pixel object. The latter is classified into three categories: zonal, linear and small encapsulated objects based on their spatial information in neighboring pixels and their spectral information. In order to determine the suitability and the effectiveness of the proposed approach, experiments are performed with a simulated and a real hyperspectral image.

Land-use/cover change, climate change, sea level evolution are examples of application that are associated with change detection. Actually, we use satellite image time series to monitor the change where entities are often dynamic along time. Moreover, knowledge associated to these spatio-temporal objects can evolve when changes occur. Thus, for modeling this kind of knowledge it is necessary to deal with four aspects: spectral, spatial, temporal and semantic. Such approach can be modeled by ontologies in many levels. Thereby, a shared ontology can be an ontology or a combination of some ontologies based on some mechanisms of linking. Such link process should maintain consistency between represented knowledge. In this paper, we propose a multi-level ontological approach for monitoring dynamics in remote sensing images. The proposed methodology aims to link our domain ontology to an upper level ontology thus enabling to represent existing change processes.

Nowadays, Normalized Difference Vegetation Index (NDVI) time-series has been successfully used in research regarding global environmental change. NDVI time series have proven to be a useful means of indicating drought-related vegetation conditions, due to their real-time coverage across the globe at relatively high spatial resolution. In this paper, we propose a method for detecting rare events in NDVI series. These events are particularly rare and infrequent which increases their complexity of detecting and analyzing them. The proposed method is based on the analysis of the random component by Jarque-Bera test to verify the presence of rare events and obtain their features (time and amplitude). For validation, we have used a database for regions in Northwestern of Tunisia. These data come from MODIS for a period from 18 February 2000 to 17 November 2013 at a spatial resolution of 250 m by 250m.

Active learning (AL) has shown a great potential in the field of remote sensing to improve the efficiency of the classification process while keeping a limited training dataset. Active learning uses heuristics to select the most informative pixels in each iteration. In literature, there are several metrics and selection criteria. In this paper, we focus on the uncertainty heuristics for large margin active learning. Existing uncertainty metrics are presented and combined to new ones using support vector machine learning algorithm. Besides, a new methodology is proposed, which automates a priori the choice of the best uncertainty heuristic. This contribution is evaluated on hyperspectral datasets while varying two parameters: class mixing and class balance. Finally discussion and conclusion are drawn.

Ontologies are considered as one of the most powerful tools for knowledge representation and reasoning. Thus, they are considered as a fundamental support for image annotation, indexing and retrieval. In order to build a remote sensing satellite image ontology that models the geographic objects that we find in a scene, their characteristics as well as their relationships, we propose to reuse existing geographic ontologies to enrich an ontological core. Reusing high quality resources (called source ontologies) helps ensuring a good quality for the extracted knowledge, and alleviating the conceptualization stage, i.e. avoiding building a new ontology from scratch. Ontology alignment is an important phase within the enrichment process. It is a process that allows discovering mappings between core and source ontologies, where each mapping is a couple of entities brought from each ontology and linked together either by an equivalence or a subsumption relationship. Such relationships are based on various similarity measures. In this paper, we first present a brief literature review of existing theoretical frameworks for similarity measures, then we describe a new alignment approach based on a semi-automatic mapping selection process that needs little human intervention. First experiments show the benefit from using the proposed approach.

Nowadays, the process of change detection is regarded as an outstanding way for urban planning and design. The major concern of this paper is to investigate the non-stationary character of multi-temporal time series. To overcome this problem, we propose an adaptive multiplicative decomposition of non-stationary multi-temporal satellite image, which allows to decompose the series into three components: trend, seasonal and random, to properly model the evolution of land cover. We carried several experiments to validate our approach based on Landsat images covering the region of “Tres Cantos-Madrid” in Spain. The obtained results show the effectiveness of our proposed method comparing to some conventional methods.

The difficulties of having expertise in expert systems, the increasing of the data volume, self adaptation and prediction, all those problems are solved in the presence of learning. The classical definition of learning in cognitive science is the ability to improve the performance as the exercise of an activity. With learning, knowledge is automatically extracted from a data set. In this paper, we are interested to study efficient active learning methods. These methods are based on the definition of an efficient training set by iteratively adapting it through adding the most informative unlabeled instances. The selection of these instances are generally based on an uncertainty and diversity criteria. This study is focused on the uncertainty criterion. A review of the principal families of active learning algorithms is presented. Then the large-margin active learning techniques are detailed and evaluations of the contribution of large margin uncertainty criteria are presented.

To be robust, decision-making process must take account the imperfection associated with models. The identification, understanding and propagation of imperfection sources are important. In general, the imperfection in land cover change (LCC) prediction process can be categorized as both aleatory and epistemic. This imperfection, which can be subdivided into parameter and structural model imperfection, is recognized to have an important impact on actual results. Previously, it has been shown that evidence theory can be applied to model aleatory and epistemic imperfection. The objective of this study is to introduce an efficient methodology for the propagation of imperfection using evidence theory in LCC prediction model, which include both parameter and structural model imperfection sources.

Hyperspectral sensors (HS) are next-generation optical sensors that have excellent spectroscopic performance with hundreds of spectral bands. Multispectral sensors (MS) are conventional optical sensors that have a few tens of spectral bands with high spatial resolution. This work aims to combine, the spectral information of the hyperspectral image with the spatial and spectral information of the multispectral image for automatic classification while considering spatial relationships between sources in low-spatial resolution data. The considered approach is validated first by using synthetic images from the USGS spectral library, but also using Hyperion sensor as hyperspectral image. And SPOT sensor as multispectral image, representing the region of Gabes Matmata in southern Tunisia.

Actually, we still have not a well established satellite image ontology that would be very useful to assist us to study major facts affecting earth, detect and monitor natural phenomena. Nevertheless, there are several domain-specific geographic ontologies that can be used as semantic resources to build such an ontology. Thus, we can start from a core satellite image ontology such as the one of Durand and enrich it using these geographic ontologies. Ontology matching is one of the principal activities in the ontology enrichment process and highly depends on the similarity measures that are considered as well as the way they are combined together in order to decide whether two concepts coming from different ontologies are alienable. In the literature, research on similarity mainly focuses on issues related to how to compute and refine similarity measures. However, few research addresses studying their dependencies and contributions in the evaluation of the overall similarity between objects to be compared. In this paper and in order to align an initial remote sensing satellite image ontology with a set of geographic ontologies, we give insights on the main similarity models as well as their associated measures. We then propose a method in order to select a reduced set of the most important similarity measures to use for the alignment. Afterwards, we present a method that can produce a ranking model that allows sorting mappings between concepts coming from two different ontologies, in a decreasing order of a global similarity score. First experimentations show that the proposed approach is promising.

The interpretation of hyperspectral imagery is an essential task for classification, changes detection and monitoring of natural phenomena. One challenge of the processing hyperspectral images, with better spectral and temporal resolution is the huge amount of data volume and the interpretation in high dimensionality data. Various techniques have been developed in the literature for dimensionality reduction, generally divided into two main categories: projection techniques and bands selection techniques. In this work, we present a new approach for interpretation in hyperspectral imagery based on hybrid dimensionality reduction methods. The presented approach combines a projection method with a bands selection method. Indeed, the objective of the research is to obtain an efficient hyperspectral image interpretation. The performances of the proposed approach were evaluated using AVIRIS hyperspectral image.

Multitemporal hyperspectral images are gaining an ever-increasing importance revealed by the ambition of the remote sensing community to develop new generation of sensors. Therefore, multitemporal images classification and change detection issues are greatly relevant in several research topics. In this paper, we propose a novel approach for modeling the temporal variation of the reflectance response as a function of time period and wavelength; summarizing the spectral signature of hyperspectral pixels as a 3-D mesh. This approach is adopted for hyperspectral time series analysis leading to the main following contribution: an advanced form of the temporal spectral signature defining the reflectance at each pixel as a congregation of the spatial/spectral/temporal dimensions. Afterward, by formulating the temporal data set in an adequate multidimensional feature space of contextual data, an innovative processing scheme exploiting the theoretical backgrounds of 3-D surface reconstruction and matching is adopted for data interpretation. Finally, an improved method for multitemporal endmember extraction and spectral unmixing based on multilinear algebra methods is introduced. A case study, in a region located in southern Tunisia, is conducted on a multitemporal subset of Hyperion images. Up to 89.86% of sampling sites have been correctly predicted by the proposed approach, outperforming conventional classifiers. The good performances obtained, on simulated multitemporal images and over various real experimental scenarios, illustrate the effectiveness and the generalization capacities of the proposed approach.

This paper develops a noise-robust processing method which can be used to enhance the classification of remotely sensed hyperspectral images. The method first illustrates the benefit of boosting the classical classifiers by exploiting the capability of belief functions. The evidential approach is adopted to produce a map which is approximately insensitive to the noise accompanying the original hyperspectral data-set. Then, a new Evidential Kernel Fisher Discriminant is proposed by using a modified version of the Expectation-Maximization (EM) algorithm. An experimental comparison of the proposed approach with other classical methods is conducted using both synthetic and real hyperspectral data collected by the HYPERION sensor. Our experiments reveal that both classification and unmixing process can benefit from the proposed aggregated approach, remarkably, when the noise level present in the original hyperspectral series is propositionally high.

Within the framework of the unmixing of hyperspectral images, the pixel mixture is a difficult problem to solve. This difficulty comes from several outliers which seriously affect the reliability of spectral unmixing results. The illumination change effect, where the image does not reflect the true appearance of the scene, in many cases due to shadow facts, is considered to be one of the most important outliers. The present work proposes a new approach called Spectral Angle Measure-based Spectral Unmixing which uses the spectral angle constraint for abundance quantification. The major benefit of this approach is its ability to take advantage of the geometric properties of the Spectral Angle Measure technique to estimate abundance quantification independently of the amplitude (magnitude) of the Endmembers spectral signatures, using only spectral angle measures. As a consequence, a significant reduction in spectral unmixed error corresponding to the spectral similarity within-class confusion is obtained. A second benefit concerns physical constraints which are respected. The experiment was conducted using simulated and real images to validate our approach and to compare it with a well known statistical one.

Satellite image classification is usually marked by several types of imperfection such as uncertainty, imprecision, and ignorance. Data fusion of additional sensors tries to overcome the types of imperfection by using probability, possibility, and evidence theories. Our approach will lead to improve classification accuracy of satellite images by choosing the optimum theory for a particular image context and proposing a theoretical framework based on a multiagent system and case-based reasoning. We validate our approach trough a set of optical images from the satellite Satellite Positioning and Tracking 4 and radar images from the European Remote Sensing satellite 2, and we show that the overall accuracy is considerably increased from 83% for maximum-likelihood classification applied to multispectral imagery to 94% with the proposed approach.

Land cover interpretation using multisensor remote sensing images is an important task that allows the extraction of information that is useful for several applications. However, satellite images are usually characterized by several types of imperfection, such as uncertainty, imprecision, and ignorance. Using additional sensors can help improve the image interpretation process and decrease the associated imperfections. Fusion methods such as the probability, possibility, and evidence methods can be used to combine information coming from these sensors. An extensive literature has accumulated during the last decade to resolve the issue of choosing the best fusion method, particularly for satellite images. In this paper, we present a semiautomatic approach based on case-based reasoning (CBR) and rule-based reasoning, allowing intelligent fusion method retrieval. This approach takes into account the advantage of data stored in the case base, allowing a more efficient processing and a decrease in image imperfections. The proposed approach incorporates three modules. The first is a learning module based on evaluating three fusion methods (probability, possibility, and evidence) applied to the given satellite images. The second looks for the best fusion method using CBR. The last is devoted to the fusion of multisensor images using the method retrieved by CBR. We validate our approach on a set of optical images coming from the Satellite Pour l'Observation de la Terre 4 and radar images coming from European Remote Sensing Satellite 2 (ERS-2) representing a central Tunisian region.

With the rapid development in remote sensing, digital image processing becomes an important tool for quantitative and statistical interpretation of remotely sensed images. These images, often, contain complex and natural scenes. The constant increase in the amount of data to treat issued from satellites images, has made automatic content extraction and retrieval highly desired goals for effective and efficient processing of remotely sensed imagery. One of the main difficulties of these applications is the knowledge representation of objects, scene and interpretation strategy. In this paper, we present an integrated hierarchical approach based on the use of a hierarchical blackboard architecture and multi-agent system in order to increase the degree of semi-automatic interpretation of remotely sensed images. This hierarchical architecture is motivated in order to avoid the bottleneck caused by the growing number of the knowledge sources on a single blackboard, reduce the information complexity and complex tasks and increase the system efficiency whenever the information is distributed over several blackboard levels. In this paper, a stage of image analysis has been examined in order to establish the viability of MAS and hierarchical blackboard architecture for change detection. A set of Spot multi-temporal images, was analyzed in terms of spectral responses from different land cover types.

Summary form only given. Geographical data are characterized by their complexity, which has resulted in the appearance of new heterogeneous GIS formats used currently in very varied applications. In this paper, we propose to design and develop a system for the management of geo-referenced data while being based on a set of generic tools treating various GIS formats and to transform them into a standardized single interchange format which could be used in several applications. For realization of this system we used tools such as the GIS GRASS, the translation library GDAL/OGR, the conversion tool FME, the standard XML development environment XMLSpy. Thus, we used languages such as the modeling language UML used to model the geographical data, GML representing the storage and exchange format of geographical data, XSL which is a stylesheet language, and SVG that is a language for the 2D graphic representation of XML data. The system suggested is characterized by a modular architecture allowing fast evolution and a rather easy development.

The number of satellite images is in full evolution allowing us to improve the extraction of useful information related to the physical reality of natural scenes. In this paper, we propose a new hybrid approach of multi-spectral analysis of satellite images. This approach consists in a method of blind separation of sources. This method allows us to decompose a pixel into information coming from independent sources. Algorithms adapted in the context of our work, operating in the two dimensional space, have been used for the separation. These algorithms produces many sources, in order to choose among them the most significant having a maximum value of entropy representing a maximum information about one class of land use. In order to have a classified image with good classification, we proceed with the fusion of these sources using a technique of maximum likelihood classification. We validated our approach on optical images of the satellite SPOT 4 and radar images of the satellite ERS 2 representing a central Tunisian region. The results obtained consists in the production of five classes of land use.

Presents the application of blind source separation (BSS) to satellite images produced with many sensors. These sensors measure the electromagnetic radiation emitted or reflected by the studied surface. Each pixel of these images represent a radiometric value that results from several objects or sources. So, a pixel is considered as a mixture of different sources. We propose to examine the ability of BSS methods to restore the independent sources by the use of algorithms based on high-order statistics such as the independent component analysis (ICA) and second order blind identification (SOBI). These techniques allow us to identify the images sources and the mixing and unmixing matrix in order to help the photointerpreter to improve his analysis of the satellite images. Each image source has maximum information about one source that can represent a class of land use. In order to have a classified image, we proceed with the fusion of these sources using a technique of maximum likelihood classification (MLC). The results obtained consist of five classes of land use (parcel region, urban region, humid region, cultivated region and sebkhat).