Search

Land cover (LC) is an essential variable for environmental monitoring in many application domains. The detection of changes in LC can support the understanding of environmental dynamics. However, LC legends present a high degree of inconsistencies that significantly reduce their usability. This study investigates the effectiveness of ISO standard 19144-2, better known as Land Cover Meta-Language (LCML), to improve the standardization and harmonization of different LC taxonomies and maps. LCML vocabulary and syntactic rules facilitate the integration of natural resources information. LC classes are represented by a sequence of “Basic Elements” and attributes defined as “Properties” and “Characteristics.” Such elements are formalized in a Unified Modeling Language class diagram. This study presents first, a method to evaluate and score the “similarity” of different LCML legends, second, an application of the similarity assessment criteria to an area located in Bangladesh for translating its specific LCML legend into a different taxonomy, i.e., the System of Environmental Economic Accounting, and third, a Python implementation to be incorporated in new or already existing tools. The results obtained show that when class similarity assessment is carried out by Basic Elements only, the process performs well for simple classes. When classes are characterized by similar basic elements (e.g., biotic elements) structure, the introduction of class properties is needed to disambiguate complex situations. The findings indicate that the proposed methodology can exploit LCML land feature semantic representation. Moreover, it can be used for translating LCML classes into different taxonomies, for facilitating class comparison and change detection.

Assessing and maintaining the conservation of natural and semi-natural grassland ecosystems is one of the most important actions of the Biodiversity Strategy by the European Commission.The present study focuses on the detection of long-term changes, from 1990 to 2018, of natural grasslands ecosystem, at local scale, in the “Murgia Alta”, a National Park as well as a Natura 2000 protected area, Southern Italy. The study site represents one of the largest areas for the conservation of such ecosystem in Italy. It is under pressure and in danger of destruction due to soil graining for agricultural intensification, illegal expansion of extraction sites, fires and land abandonment with consequent biodiversity loss.Land Cover (LC) changes and class trends are one of the measures (sub-indicator) required for the implementation of the Sustainable Development Goals (SDG) 15.3.1 Indicator (“Proportion of land that is degraded over total land area”) of the Agenda 2030 by United Nations.Multisource/multiresolution free available satellite data (visible, near infrared and short wave infrared spectral bands) were considered. Historical images from Landsat (4 images per year, one per season) were analyzed to produce different LC multiclass maps for 1990, 2001, 2004, 2011 and 2018 at 30 m spatial resolution, with an automatic data-driven classifier (Support Vector Machine). For 2018 Sentinel-2 data, 10 m spatial resolution, were also considered.The mean value of the Overall Accuracies obtained for the LC maps from Landsat was 95%. Similar value was obtained in the last year from Sentinel-2.Then natural grassland layer was extracted from those maps to analyze the trend of the grasslands ecosystem over time. The findings obtained indicate a total loss in the extension of the ecosystem of about 18% from 1990 to 2018. The major decrease (26%) occurred in 1990-2001. Then a modest decrease followed up to 2004 (year of institution of the National Park). Finally a slight increase probably due to land abandonment followed to fire events was quantified after 2004.From the comparison of the different LC maps obtained, the decrease of natural grasslands resulted mainly due to transformation into agricultural areas.In addition, these results are consistent with those obtained using Corine Land Cover maps available for the same period although at a coarser scale.The SDG sub-indicator was evaluated inside the protected area and in a buffer, 10 km, area around. This sub-measure, which can be evaluated from time-series of satellite free data, can support long-term monitoring of protected area and can be used not only for the resilience evaluation of the study site to climate changes but also for the evaluation of conservation policies and as input to scenario modelling.

The present study proposes a method based on the integration of EO and census data for updating the spatial distribution map of the migrant population regularly residing in an area very exposed to migration on the Southern Mediterranean coast, i.e. Bari (Italy) and its surrounding areas. The method implemented is a vector-based variant of the dasymetric mapping approach used by the Joint Research Center (JRC) within the Data for Integration initiative to disaggregate data from census areas into a uniform grid and it draws on updated census data (1st January 2019) concerning regular migrant population, an updated human settlements (built-up) map and information about their use. The mapping of the built-up areas results from either a data-driven or knowledge-driven automatic classification techniques of multi-temporal Sentinel-2 satellite images acquired during 2018 with 10 m spatial resolution. The spatial distribution (density) map of regular migrant population obtained along with other updated ancillary data were used, at local (intra-urban) scale, to provide Sustainable Development Goal (SDG) 11 indicators. These include the ratio of land consumption rate to regular migrant population growth rate (SDG 11.3.1) and the map of regular migrant population living in inadequate housing (SDG 11.1.1) as specific categories of SDG 11 for the epochs 2011 and 2018. At the local level, the regular migrant population density map and the SDG 11 indicator values were provided for each cell of a 100 m by 100 m output grid. In recent literature, not many authors have tried to implement SDG 11 indicators at intra-urban scale. The findings obtained in the study area indicate an increase of migrant population up to 44.5% in 2018 with respect to the previous estimate provided by JRC and based on 2011 census data. In addition, the results reveal the tendency by the communities to aggregate in specific neighborhoods located in central areas of the city characterized by urban decay and abandoned buildings. The products obtained through the dasymetric method applied could be useful to support urban planners and decision makers not only in the management of increasing migration pressure but also for monitoring the progress in Agenda 2030 related to SDG 11 indicators at local level.

Summary 1.Long-term habitat mapping and change detection are essential for the management of coastal wetlands as well as for evaluating the impact of conservation policies. Earth Observation (EO) data and techniques are a valuable resource for long-term habitat mapping. Although the use of EO data is well developed for the automatic production of Land Cover (LC) maps this is not the same for habitat maps, which are highly related to biodiversity. 2.In a previous paper, we used the Food and Agricultural Organization (FAO) Land Cover Classification System (LCCS) environmental attributes (e.g. water quality, lithology, soil surface aspect) for LC-to-habitat class translation. However, these environmental attributes are often not openly available, not updated or are missing. 3.This paper offers an alternative, knowledge-based solution to automatic habitat mapping. When only expert rules and EO data are used, the final overall map accuracy which is obtained by comparing reference ground-truth patches to the ones depicted in the output map, is lower (75.1%) than the accuracy obtained using environmental attributes alone (97.0%). Some ambiguities that still remain in habitat discrimination are resolved by integrating the use of LCCS environmental attributes (if available), and expert rules. 4.In this paper we use Very High Resolution (VHR) satellite data and LIDAR data. LC classes are labelled according to the LCCS taxonomy, which offers a framework to integrate EO data with in situ and ancillary data. Output habitat classes are labelled according to the European Habitats Directive (92/43 EEC Directive) Annex I habitat types and Eunis habitat classification. Two Natura 2000 coastal wetland sites in southern Italy are considered. 5.Synthesis and applications. In this paper we study the exploitation of ecological rules on vegetation pattern, plant phenology and habitat geometric properties for automatic translation of Land Cover (LC) maps to habitat maps in coastal wetlands. The methodology is useful for relatively inaccessible sites (e.g. wetlands) as it does not require in-field campaigns (generally costly) but only the elicitation of ecological expert rules. This can support site (e.g. Natura 2000) managers in long-term automatic habitat mapping. Habitat changes can be automatically detected by comparing map pairs and trends can be quantified. This is particularly useful to satisfy the commitments of the European Habitats Directive (92/43/EEC), which requires Member States to take measures to maintain as, or restore to, favourable conservation status those natural habitat types and species of community interest that are listed in the Annexes to the Directive.

Through the EU FP7 Horizon 2020 ECOPOTENTIAL project, a novel and innovative approach to classification has been developed, which is termed the Earth Observation Data for Ecosystem Monitoring (EODESM), and has been built on concepts behind an implementation of the Earth Observation Data for Habitat Monitoring (EODHaM) system generated as part of the EU FP7 BIOSOS project, applied to Very High Resolution (VHR) W orldview data. The EODESM system facilitates routine classification of land covers according to the Food and Agricultural Organisations Land Cover Classification System (FAO-LCCS), translates these to other taxonomies (including General Habitat Classifications; GHCs) and facilitates routine detection of change and the generation of maps indicating the causes and consequences of such change.

There is an increasing need of effective monitoring systems for habitat quality assessment. Methods based on remote sensing (RS) features, such as vegetation indices, have been proposed as promising approaches, complementing methods based on categorical data to support decision making. Here, we evaluate the ability of Earth observation (EO) data, based on a new automated, knowledge-driven system, to predict several indicators for oak woodland habitat quality in a Portuguese Natura 2000 site. We collected in-field data on five habitat quality indicators in vegetation plots from woodland habitats of a landscape undergoing agricultural abandonment. Forty-three predictors were calculated, and a multi-model inference framework was applied to evaluate the predictive strength of each data set for the several quality indicators. Three indicators were mainly explained by predictors related to landscape and neighbourhood structure. Overall, competing models based on the products of the automated knowledge-driven system had the best performance to explain quality indicators, compared to models based on manually classified land cover data. The system outputs in terms of both land cover classes and spectral/landscape indices were considered in the study, which highlights the advantages of combining EO data with RS techniques and improved modelling based on sound ecological hypotheses. Our findings strongly suggest that some features of habitat quality, such as structure and habitat composition, can be effectively monitored from EO data combined with in-field campaigns as part of an integrative monitoring framework for habitat status assessment.

Traditionally, analyses of relationships between amphibians and habitat focused on breeding environments (i.e., pond features) more than on the features of the surrounding environment. Nevertheless, for most amphibians the terrestrial phase is longer than the aquatic phase, and consequently landscape features (i.e., habitat mosaics) may have an important role for modelling amphibian distribution. There were different aims in this analysis. Firstly, we compared the effectiveness of the information provided by land cover/use (LC/LU) classes and habitat classes defined according to a new habitat taxonomy named General Habitat Category (GHC), which is based on the concept of biological forms of dominant vegetation and class naturalness. The GHC map used was obtained from a pre-existing validated LC/LU map, by integrating spectral and spatial measurements from very high resolution Earth observation data according to ecological expert rules involving concepts related to spatial and temporal relationships among LC/LU and habitat classes. Then, we investigated the importance for amphibians of the landscape surrounding ponds within the Italian Alta Murgia National Park. The work assessed whether LC/LU classes in pond surrounds are important for the presence/absence of amphibians in this area, and identified which classes are more important for amphibians. The results obtained can provide useful indications to management strategies aiming at the conservation of amphibians within the study area. An information-theoretic approach was adopted to assess whether GHC maps allow to improve the performance of species distribution models. We used the Akaike's Information Criterion (AICc) to compare the effectiveness of GHC categories versus LC/LU categories in explaining the presence/absence of pool frogs. AICc weights suggest that GHC categories can better explain the distribution of frogs, compared to LC/LU classes.

Periodic monitoring of biodiversity changes at a landscape scale constitutes a key issue for conservation managers. Earth Observation (EO) data offers a potential solution, through direct or indirect mapping of species or habitats. Most national and international programs rely on the use of Land Cover (LC) and/or Land Use (LU) classification systems. Yet, these are not as clearly relatable to biodiversity in comparison to habitat classifications, and provide less scope for monitoring. While a conversion from LC/LU classification to habitat classification can be of great utility, differences in definitions and criteria have so far limited the establishment of a unified approach for such translation between these two classification systems. Focusing on five Mediterranean NATURA 2000 sites, this paper considers the scope for three of the most commonly used global LC/LU taxonomies - CORINE Land Cover (CLC), the Food and Agricultural Organisation (FAO) Land Cover Classification System (LCCS) and the International Geosphere-Biosphere Programme (IGBP) to be translated to habitat taxonomies. Through both quantitative and expert knowledge based qualitative analysis of selected taxonomies, FAO-LCCS turns out to be the best candidate to cope with the complexity of habitat description and provides a framework for EO and in-situ data integration for habitat mapping, reducing uncertainties and class overlaps and bridging the gap between LC/LU and habitats domains for landscape monitoring - a major issue for conservation. This study also highlights the need to modify the FAO-LCCS hierarchical class description process to permit the addition of attributes based on class-specific expert knowledge to select multi-temporal (seasonal) EO data and improve classification. An application of LC/LU to habitat mapping is provided for a coastal Natura 2000 site with high classification accuracy as a result Key words: Mapping; land cover; land use; habitat; earth observation; taxonomies; Natura 2000; classification schemes

Protected areas are experiencing increased levels of human pressure. To enable appropriate conservation action, it is critical to map and monitor changes in the type and extent of land cover/use and habitat classes, which can be related to human pressures over time. Satellite Earth observation (EO) data and techniques offer the opportunity to detect such changes. Yet association with field information and expert interpretation by ecologists is required to interpret, qualify and link these changes to human pressure. There is thus an urgent need to harmonize the technical background of experts in the field of EO data analysis with the terminology of ecologists, protected area management authorities and policy makers in order to provide meaningful, context-specific value-added EO products. This paper builds on the DPSIR framework, providing a terminology to relate the concepts of state, pressures, and drivers with the application of EO analysis. The type of pressure can be inferred through the detection of changes in state (i.e. changes in land cover and/or habitat type and/or condition). Four broad categories of changes in state are identified, i.e. land cover/habitat conversion, land cover/habitat modification, habitat fragmentation and changes in landscape connectivity, and changes in plant community structure. These categories of change in state can be mapped through EO analyses, with the goal of using expert judgement to relate changes in state to causal direct anthropogenic pressures. Drawing on expert knowledge, a set of protected areas located in diverse socio-ecological contexts and subject to a variety of pressures are analysed to (a) link the four categories of changes in state of land cover/habitats to the drivers (anthropogenic pressure), as relevant to specific target land cover and habitat classes; (b) identify (for pressure mapping) the most appropriate spatial and temporal EO data sources as well as interpretations from ecologists and field data useful in connection with EO data analysis. We provide detailed examples for two protected areas, demonstrating the use of EO data for detection of land cover/habitat change, coupled with expert interpretation to relate such change to specific anthropogenic pressures. We conclude with a discussion of the limitations and feasibility of using EO data and techniques to identify anthropogenic pressures, suggesting additional research efforts required in this direction.

Although underestimated in practice, the small/unrepresentative sample problem is likely to affect a large segment of real-world remotely sensed (RS) image mapping applications where ground truth knowledge is typically expensive, tedious, or difficult to gather. Starting from this realistic assumption, subjective (weak) but ample evidence of the relative effectiveness of existing unsupervised and supervised data labeling systems is collected in two RS image classification problems. To provide a fair assessment of competing techniques, first the two selected image datasets feature different degrees of image fragmentation and range from poorly to ill-posed. Second, different initialization strategies are tested to pass on to the mapping system at hand the maximally informative representation of prior (ground truth) knowledge. For estimating and comparing the competing systems in terms of learning ability, generalization capability, and computational efficiency when little prior knowledge is available, the recently published data-driven map quality assessment (DAMA) strategy, which is capable of capturing genuine, but small, image details in multiple reference cluster maps, is adopted in combination with a traditional resubstitution method. Collected quantitative results yield conclusions about the potential utility of the alternative techniques that appear to be realistic and useful in practice, in line with theoretical expectations and the qualitative assessment of mapping results by expert photointerpreters.

This work focuses on two challenging types of problems related to quality assessment and comparison of thematic maps generated from remote sensing (RS) images when little or no ground truth knowledge is available. These problems occur when: (1) competing thematic maps, generated from the same input RS image, assumed to be available, must be compared, but no ground truth knowledge is found to assess the accuracy of the mapping problem at hand, and (2) the generalization capability of competing classifiers must be estimated and compared when the small/unrepresentative ground truth problem affects the RS inductive learning application at hand. Specifically focused on badly posed image classification tasks, this paper presents an original data-driven (i.e., unsupervised) thematic map quality assessment (DAMA) strategy complementary (not alternative) in nature to traditional supervised map accuracy assessment techniques, driven by the expensive and error-prone digitization of ground truth knowledge. To compensate for the lack of supervised regions of interest, DAMA generates so-called multiple reference cluster maps from several blocks of the input RS image that are clustered separately. Due to the unsupervised (i.e., subjective) nature (ill-posedness) of data clustering, DAMA provides no (absolute) map accuracy measure. Rather, DAMA's map quality indexes are to be considered unsupervised (i.e., subjective) relative estimates of labeling and segmentation consistency between every competing map at hand and the set of multiple reference cluster maps. In two badly posed RS image mapping experiments, DAMA's map quality measures are proven to be: (1) useful in the relative comparison of competing mapping systems; (2) consistent with theoretical expectations; and (3) in line with mapping quality criteria adopted by expert photointerpreters. Documented limitations of DAMA are that it is intrinsically heuristic due to the subjective nature of the clustering problem, and like any evaluation measure, it cannot be injective.

The derivation of habitat maps is enhanced if land cover maps are used as basis for the mapping procedure. In this study, a supervised learning framework is proposed to perform object-based classification to General Habitat Categories. A Land Cover Classification System map is used as basis, and an approach to generate numerical features from the object land cover class names and attributes is introduced. An additional number of spectral, morphological, and topological features are extracted from very high resolution satellite imagery and classification accuracies up to 80.4% for 14 classes are reached. Inclusion of LiDAR (Light Detection And Ranging) data or proposed texture analysis features, improve accuracies to 86% and around 83%, respectively, with the latter proving as promising surrogates of LiDAR data features. The method outperformed rule-based approaches, indicating its potential in accurate and labor- and time-efficient habitat classification.

Monitoring land cover and habitat change is a key issue for conservation managers because of its potential negative impact on biodiversity. The Land Cover Classification System (LCCS) and the General Habitat Categories (GHC) System have been proposed by the remote sensing and ecological research community, respectively, for the classification of land covers and habitats across various scales. Linking the two systems can be a major step forward towards biodiversity monitoring using remote sensing. The translation between the two systems has proved to be challenging, largely because of differences in definitions and related difficulties in creating one-to-one relationships between the two systems. This paper proposes a system of rules for linking the two systems and additionally identifies requirements for site-specific contextual and environmental information to enable the translation. As an illustration, the LCCS classification of the Le Cesine protected area in Italy is used to show rules for translating the LCCS classes to GHCs. This study demonstrates the benefits of a translation system for biodiversity monitoring using remote sensing data but also shows that a successful translation is often depending on the degree of ecological knowledge of the habitats and its relationship with land cover and contextual information.

Habitat mapping is a core element in numerous tasks related to sustainability management, conservation planning and biodiversity monitoring. Land cover classifications, extracted in a timely and area-extensive manner through remote sensing data, can be employed to derive habitat maps, through the use of domain expert knowledge and ancillary information. However, complete information to fully discriminate habitat classes is rarely available, while expert knowledge may suffer from uncertainty and inaccuracies. In this study, a rule-based classification methodology for habitat mapping through the use of a pre-existing land cover map and remote sensing data is proposed to deal with uncertainty, missing information, noise afflicted data and inaccurate rule thresholds. The use of the Dempster–Shafer theory of evidence is introduced in land cover to habitat mapping, in combination with fuzzy logic. The framework is able to handle lack of information, by considering composite classes, when necessary data for the discrimination of the constituting single classes is missing, and deal with uncertainty expressed in domain expert knowledge. In addition, a number of fuzzification schemes are proposed to be incorporated in the methodology in order to increase its performance and robustness towards noise afflicted data or inaccurate rule thresholds. Comparison with reference data reveals the improved performance of the methodology and the efficient handling of uncertainty in expert rules. The further scope is to provide a robust methodology readily transferable and applicable to similar sites in different geographic regions and environments. Although developed for habitat mapping, the proposed rule-based methodology is flexible and generic and may be well extended and applied in various classification tasks, aiming at handling uncertainty, missing information and inaccuracies in data or expert rules.

This paper presents a method for hedgerow extraction from very high resolution optical images using image segmentation. The method is based on a connected component region growing approach followed by an object based image analysis filtering allowing to introduce high level knowledge about the characteristics of the desired hedgerows. The proposed approach is used as a part of the EODHaM image processing system developed in the framework of the BIO SOS FP7 project.

One of the core European Union environmental policies is the creation and monitoring of the Natura 2000 network of protected areas. This network has been explicitly established for the preservation of conservation priority habitat types and species. Still the concept of habitat is a key concept for ecologists that remains ill defined and is notoriously hard to quantify and measure. Several classification schemes have been put forward, but their relative strengths and weaknesses remain less well examined. In this study we analyzed 8 different Natura 2000 sites (3 Greek, 2 Italian, 2 Portuguese, 1 British). Our study sites reflect a variety of ecosystems, most of them are Mediterranean (7 of the 8) and most of them are wetlands (6 of the 8). In each site, we classified habitats according to 4 different classification schemes (Annex I of the Habitats Directive, Corine Biotopes, EUNIS and General Habitat categories). Also, we used three other widely used land cover classification schemes (namely Corine Land Cover, FAO Land Cover Classification System and IGBP DIS scheme). We found that the different schemes produced considerably different values of landscape diversity leading even to different ranking of the sites according to their diversity. Furthermore, when comparing the landscape composition among sites according to the different schemes, they led to different inferences. Our results imply that the classification scheme used for estimating habitat composition plays an important role for the monitoring of protected areas, perhaps more important than previously assumed.

The EODHaM system offers the right mix of site specific configurability and pre-engineered processing modules and workflows, which allow to offer a habitat monitoring service, particularly suited to support the multi-annual monitoring of Natura 2000 sites. The service provisioning foresees for each site an inception phase, during which the service provider collaborates with domain experts from the site in order to setup together the optimal service chain and processing configuration, customizing the workflow steps from the range of prebuilt, configurable processing modules.

Focusing on the Food and Agricultural Organization (FAO) Land Cover Classification System (LCCS) and the recently proposed General Habitat Categories (GHCs) classifycation system, this paper illustrates how expert knowledge concerning class spatial arrangement in the scene at hand class, class phenology and class spectral signature in multitemporal EO images can fill the gaps between the two classification systems and provide LC/LU to habitat translation. An application to a Natura 2000 site in Southern Italy which includes a wetland costal area is discussed.

The BIO_SOS project (www.biosos.eu) focuses on developing a robust system for habitat mapping and monitoring, named EODHaM, with this based on prior generation of land cover/use (LC/LU) and change map maps and their subsequent translation to categories of habitat appropriate to support conservation agencies and land managers in decisions relating to protection of Natura 2000 sites. The input data sources are multi-seasonal EO measurements and on-site data, including ancillary information and in-field measurements. For this purpose, the Food and Agricultural Organisation (FAO) Land Cover Classification Scheme (LCCS) and the General Habitat Categories (GHCs), from which Annex I Habitats can be defined, have been proposed for describing LC/LU and habitat categories.

A rule-based classification system for images remotely sensed by satellites is presented. The spectral characterization of the land-cover classes studied has been carried out by exploiting rules which are based on the knowledge of an expert agronomist, and applying them to a multitemporal Thematic Mapper image dataset. The system knowledge base consists of a set of rules describing each class on each date. In the construction of these rules the system adopts linguistic descriptors such as low, high, very low, and any logical combination of these predicates as the user interface. These linguistic predicates are treated as membership functions for ‘fuzzy sets’ defined on the image spectral values. The study has been carried out on a hilly area in Southern Italy which features a great variety both in terms of land use classes and in terms of agricultural practices. The following I/II level classes have been investigated: cropland, forestry, vineyards, olive-groves, pasture, bare soils and urban area...

—This paper presents a fuzzy expert system for the creation of landslide possibility maps using change of land-use data from Earth observation, as well as historical, rainfall, and earthquake data stored in a geographic information system, as input. The difference with other systems is in the use of change (differential) input data. The method is tested with 16 documented landslides. The fuzzy neural network (NN) developed can predict the crowns of 13 out of the 16 landslides to be among the 5%most at-risk pixels that are identified in the area of study, which covers 100 km2. The fuzzy expert system considers the rules that increase the possibility of a landslide, as supplied by experts, and expresses them in the form of an empirical algebraic formula. It then fuzzifies the various thresholds they rely on and, in conjunction with uncertainties that are reported by the classifier that decides the land-use change, produces a fuzzy algebraic formula that may be used to identify the range of uncertainty in the possibility of a landslide in terms of the ranges of uncertainty in the input variables. This formula is used to train an Ishibuchi fuzzy NN, which has been designed to capture uncertainty in the rules and uncertainty in the input variables. It is this Ishibuchi NN that acts as a fuzzy expert system.

This paper deals with the problem of badly posed image classification. Although underestimated in practice, bad-posedness is likely to affect many real-world image classification tasks, where reference samples are difficult to collect (e.g., in remote sensing (RS) image mapping) and/or spatial autocorrelation is relevant. In an image classification context affected by a lack of reference samples, an original inductive learning multiscale image classifier, termed multiscale semisupervised expectation maximization (MSEM), is proposed. The rationale behind MSEM is to combine useful complementary properties of two alternative data mapping procedures recently published outside of image processing literature, namely, the multiscale modified Pappas adaptive clustering (MPAC) algorithm and the sample-based semisupervised expectation maximization (SEM) classifier. To demonstrate its potential utility, MSEM is compared against nonstandard classifiers, such as MPAC, SEM and the single-scale contextual SEM (CSEM) classifier, besides against well-known standard classifiers in two RS image classification problems featuring few reference samples and modestly useful texture information. These experiments yield weak (subjective) but numerous quantitative map quality indexes that are consistent with both theoretical considerations and qualitative evaluations by expert photointerpreters. According to these quantitative results, MSEM is competitive in terms of overall image mapping performance at the cost of a computational overhead three to six times superior to that of its most interesting rival, SEM. More in general, our experiments confirm that, even if they rely on heavy class-conditional normal distribution assumptions that may not be true in many real-world problems (e.g., in highly textured images), semisupervised classifiers based on the iterative expectation maximization Gaussian mixture model solution can be very powerful in practice when: 1) there is a lack of reference samples with respect to the problem/model complexity and 2) texture information is considered negligible (i.e., a piecewise constant image model holds).

Based on purely spectral-domain prior knowledge taken from the remote sensing (RS) literature, an original spectral (fuzzy) rule-based per-pixel classifier is proposed. Requiring no training and supervision to run, the proposed spectral rule-based system is suitable for the preliminary classification (primal sketch, in the Marr sense) of Landsat-5 Thematic Mapper and Landsat-7 Enhanced Thematic Mapper Plus images calibrated into planetary reflectance (albedo) and at-satellite temperature. The classification system consists of a modular hierarchical top-down processing structure, which is adaptive to image statistics, computationally efficient, and easy to modify, augment, or scale to other sensors' spectral properties, like those of the Advanced Spaceborne Thermal Emission and Reflection Radiometer and of the Satellite Pour l'Observation de la Terre (SPOT-4 and -5). As output, the proposed system detects a set of meaningful and reliable fuzzy spectral layers (strata) consistent (in terms of one-to-one or many-to-one relationships) with land cover classes found in levels I and II of the U.S. Geological Survey classification scheme. Although kernel spectral categories (e.g., strong vegetation) are detected without requiring any reference sample, their symbolic meaning is intermediate between those (low) of clusters and segments and those (high) of land cover classes (e.g., forest). This means that the application domain of the kernel spectral strata is by no means alternative to RS data clustering, image segmentation, and land cover classification. Rather, prior knowledge-based kernel spectral categories are naturally suitable for driving stratified application-specific classification, clustering, or segmentation of RS imagery that could involve training and supervision. The efficacy and robustness of the proposed rule-based system are tested in two operational RS image classification problems. © 2006 IEEE.

Radial Basis Function (RBF) classifiers, which consist of an hidden and an output layer, are traditionally trained with a two-stage hybrid learning approach. This approach combines an unsupervised (data-driven) first stage to adapt RBF hidden layer parameters with a supervised (error-driven) second stage to learn RBF output weights. Several simple strategies that exploit labeled data in the adaptation of centers and spread parameters of RBF hidden units may be pursued. Some of these strategies have been shown to reduce traditional weaknesses of RBF classificaton, while typical advantages are maintained, e.g., fast training, easy implementation, low responses to outliers. In this work, we compare a traditional RBF two-stage hybrid learning procedure with an RBF two-stage learning technique exploiting labeled data to adapt hidden unit parameters. Two data sets were analized: the first consisted of multitemporal remote sensed data; the second consisted of Magnetic Resonance images...

In recent years it has been proved that combined analysis of SAR intensity and interferometric correlation images is a valuable tool in classification tasks where traditional techniques such as crisp thresholding schemes and classical maximum likelihood classifiers have been employed. In this work, developed in the framework of the ESA AO3-320 project titled Application of ERS data to landslide activity monitoring in southern Apennines, Italy, our goal is to investigate: (1) usefulness of SAR interferometric correlation information in mapping areas with diffuse erosional activity, including landslides; and (2) effectiveness of soft computing techniques in the combined analysis of SAR intensity and interferometric correlation images. Two neural classifiers are selected from the literature. The first classifier is a one- stage error-driven Multilayer Perceptron (MLP) and the second classifier is a Two-Stage Hybrid (TSH) learning system, consisting of a sequence of an unsupervised data-driven first stage with a supervised error-driven second stage. The TSH unsupervised first stage is implemented as either: (1) the on- line learning, dynamic-sizing, dynamic-linking Fully Self Organizing Simplified Adaptive Resonance Theory (FOSART) clustering model; (2) the batch-learning, static-sizing, no- linking Fuzzy Learning Vector Quantization (FLVQ) algorithm; or (3) the on-line learning, static-sizing, static-linking Self-Organizing Map (SOM). The input data set consists of three SAR ERS-1/ERS-2 tandem pair images depicting an area featuring slope instability phenomena in the Campanian Apennines of Southern Italy. From each tandem pair, four pixel-based features are extracted: the backscattering mean intensity, the interferometric coherence, the backscattering intensity texture and the backscattering intensity change. Our classification task is focused on the discrimination of land cover types useful for hazard evaluation, i.e., evaluation of areas affected by erosion. Classification results show that class erosion can be discriminated from other land cover classes when SAR mean intensity images are combined with coherence and texture information. In addition, our results demonstrate that soft computing techniques provide useful tools for the combined analysis of SAR intensity and coherence images. In particular, the TSH classifier employing the FOSART clustering algorithm shows: (1) an overall accuracy comparable with that of the other classification schemes under testing; (2) a training cost significantly lower than that of MLP and lower than that of TSH employing either FLVQ or SOM as its first stage; and (3) a capability of discriminating class erosion superior to that of the other classification schemes under testing.

The objective of this paper is to discuss a state-of-the-art of methodology and algorithms for integrating fuzzy sets and neural networks in a unique framework for dealing with pattern recognition problems, in particular classification procedures. Methods of pattern recognition are studied in two main streams, namely supervised and unsupervised learning. We propose our own definition of fuzzy neural integrated networks. This criterion is proposed as a unifying framework for comparison of algorithms. In the first part of the this paper, classification methods based on rule sets or numerical data are reviewed, together with specific methods for handling classification in image processing. In the second part of this paper, several fuzzy neural clustering models are reviewed and compared. These models are: i) Self-Organizing Map (SOM); ii) Fuzzy Learning Vector Quantization (FLVQ); iii) Carpenter-Grossberg- Rosen Fuzzy Adaptive Resonance Theory (CGR Fuzzy ART); iv) Growing Neural Gas (GNG); and v) Fully self-Organizing Simplified Adaptive Resonance Theory (FOSART).

This study investigates the applicability of a multimodular neuro-fuzzy system in the multispectral analysis of magnetic resonance (MR) images of the human brain. The system consists of two components: an unsupervised neural module for image segmentation in tissue regions and a supervised module for tissue labeling. The former is the fuzzy Kohonen clustering network (FKCN). The latter is a feed-forward network based on the back-propagation learning rule. The results obtained with the FKCN have been compared with those extracted by a self organizing map (SOM). The system has been used to analyze the multispectral MR brain images of a healthy volunteer. The data set included the proton density (PD), T2, T1 weighted spin-echo (SE) bands and a new T1- weighted three dimensional sequence, i.e. the magnetization- prepared rapid gradient echo (MP-RAGE). One of the main objectives of this study has been to evaluate the usefulness of brain imaging with the MP-RAGE sequence in view of automatic tissue classification. To this purpose, a quantitative evaluation has been provided on the base of some labeled areas selected interactively by a neuro- radiologist from the input raw images. Quantitative results seem to indicate that the MP-RAGE sequence may provide higher tissue separability than the T1-weighted SE sequence.

In this work the effectiveness of the fuzzy Kohonen clustering network (FKCN) has been explored in two classification experiments of remote sensed data. The FKCN has been introduced in a multi-modular neural classification system for feature extraction before labeling. The unsupervised module is connected in cascade with the next supervised module, based on the backpropagation learning rule. The performance of the FKCN has been evaluated in comparison with those of a conventional Kohonen self organizing map (SOM) neural network. Experimental results have proved that the fuzzy clustering network can be used for complex data pre-processing.

In this paper a modular neural network architecture is proposed for classification of Remote Sensed data. The neural network learning task of the supervised Multi Layer Perceptron (MLP) Classifier has been made more efficient by pre-processing the input with an unsupervised feature discovery neural module. Two classification experiments have been carried for coping with two different situations, very usual in real remote sensing applications: the availability of complex data, such as high dimensional and multisourced data, and on the contrary, the case of imperfect low dimensional data set, with a limited number of samples. In the first experiment on a multitemporal data set, the Linear Propagation Network (LPN) has been introduced to evaluate the effectiveness of neural data compression stage before classification. In the second experiment on a poor data set, the Kohonen Self Organising Feature Map (SOM) Network has been introduced for clustering data before labelling. In the paper is also illustrated the criterion for the selection of an optimal number of cluster centres to be used as node number of the output SOM layer. The results of the two experiments have confirmed that modular learning performs better than the non-modular one in learning quality and speed.

[1] K. J. Hunt, R. Hass, and R. Murray-Smith, “Extending the functional equivalence of radial basis function networks and fuzzy inference systems,” IEEE Trans. Neural Networks, vol. 7, pp. 776–781, May 1996. [2] J.-S. R. Jang, C.-T. Sun, and E. Mizutani, Neuro-Fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence, MATLAB Curriculum Series. Upper Saddle River, NJ: Prentice-Hall, 1997. [3] D. F. Specht, “A generalized regression neural network,” IEEE Trans. Neural Networks, vol. 2, pp. 568–576, Nov. 1991. Errata to “Model Transitions in Descending FLVQ”

In this work, unsupervised change detection techniques, based on three different way to compare images, are presented. Two Landsat TM registered and corrected multi-spectral images, acquired on the same geographical area on 18 May 1996 and 21 May 1997, have been used. In the first comparison technique, for each pair of corresponding pixels, the spectral change vector has been computed as the squared difference in the features vectors at the two times. In the second method, the difference image has been computed using, pixel by pixel, a chi square transformation. The third technique is based on the application of a Self-Organizing Map (SOM) neural network to clusterize the two images before comparison. The three obtained difference images has been then analyzed by using a fully automatic thresholding method exploiting the expectation-maximization (EM) algorithm. The experimental results obtained for the three difference images are comparable, showing a reliable robustness of the unsupervised approach, and only few change are detected on the analyzed scene. Moreover, the experimental results have been compared with a change detection map computed by using a supervised technique, obtaining a good agreement between unsupervised and supervised results that confirms the reliability of the considered approach. The encouraging obtained results allow to use the so-computed percentage value of changes as probability of class transitions in input to a Bayesian supervised change detection method, as presented in a companion paper by the same authors. In this framework, the unsupervised approach may be used to support supervised techniques, providing land cover transitions that can be used as guess values

This paper presents a comparative evaluation between two classification strategies for the analysis of remote sensed data. The first is based on the combination of the outputs of a neural network (NN) ensemble, the second concerns the application of Support Vector Machine (SVM) classifiers. Emphasis is given to the understanding of the limits and the advantages of the two strategies to design a classifier system able to provide high generalization capability. Two sets of experiments have been carried out to classify benchmark remote sensed data. In the first set a Fuzzy Integral has been used to combine the outputs of neural classifiers in an ensemble. In the second set of experiments SVM classifiers have been trained and tested on the same data set. The comparative analysis evidences that SVM classifiers outperform an ensemble of classifiers, whose partial results are combined by a Fuzzy Integral. The training complexity of SVMs seems, however, to be a limitation to the extensive use of SVMs in complex multisource-multitemporal data classification. © Springer-Verlag Berlin Heidelberg 2006.

Radial basis function (RBF) classifiers, which consist of an hidden and an output layer, are traditionally trained with a two-stage hybrid learning approach. This approach combines an unsupervised (data-driven) first stage to adapt RBF hidden layer parameters with a supervised (error-driven) second stage to learn RBF output weights. Several simple strategies that exploit labeled data in the adaptation of centers and spread parameters of RBF hidden units may be pursued. Some of these strategies have been shown to reduce traditional weaknesses of RBF classification, while typical advantages are maintained. In the field of remotely sensed image classification, the authors compare a traditional RBF two-stage hybrid learning procedure with an RBF two-stage learning technique exploiting labeled data to adapt hidden unit parameters.

Single and repeat-pass SAR interferometry appears to be a new promising tool for a wide variety of applications. Recent studies have already shown that interferometric correlation, e.g. coherence and intensity images can be coupled to improve thematic land classification. The techniques used to combine and analyze such information are mainly based on thresholding schemes and classical maximum likelihood, although SAR data are not normally distributed. The main objective of this work is the validation of soft computing techniques for the analysis of ERS SAR intensity images combined with interferometric correlation images. A multi-modular neural system, composed by unsupervised and supervised architectures was applied to the analysis of one tandem ERS 1/2, pair, 27-28 August 1995, and one single ERS 2 image, 24 July 1995.

The present study focuses on the use of Convolutional Neural Networks (CNN or ConvNet) to classify a multi-seasonal dataset of Sentinel-2 images to discriminate four grassland habitats in the “Murgia Alta” protected site. To this end, we compared two approaches differing only by the first layer machinery, which, in one case, is instantiated as a fully-connected layer and, in the other case, results in a ConvNet equipped with kernels covering the whole input (wide-kernel ConvNet). A patchwise approach, tessellating training reference data in square patches, was adopted. Besides assessing the effectiveness of ConvNets with patched multispectral data, we analyzed how the information needed for classification spreads to patterns over convex sets of pixels. Our results show that: (a) with an F1-score of around 97% (5 × 5 patch size), ConvNets provides an excellent tool for patch-based pattern recognition with multispectral input data without requiring special feature extraction; (b) the information spreads over the limit of a single pixel: the performance of the network increases until 5 × 5 patch sizes are used and then ConvNet performance starts decreasing.