Object-oriented Image Analysis Research Articles

Object-oriented mapping of landslides using Random Forests

Landslide inventory mapping is an indispensable prerequisite for reliable hazard and risk analysis, and with the increasing availability of very high resolution (VHR) remote sensing imagery the creation and updating of such inventories on regular bases and directly after major events is becoming possible. The diversity of landslide processes and spectral similarities of affected areas with other landscape elements pose major challenges for automated image processing, and time-consuming manual image interpretation and field surveys are still the most commonly applied mapping techniques. Taking advantage of recent advances in object-oriented image analysis (OOA) and machine learning algorithms, a supervised workflow is proposed in this study to reduce manual labor and objectify the choice of significant object features and classification thresholds. A sequence of image segmentation, feature selection, object classification and error balancing was developed and tested on a variety of sample datasets (Quickbird, IKONOS, Geoeye-1, aerial photographs) of four sites in the northern hemisphere recently affected by landslides (Haiti, Italy, China, France). Besides object metrics, such as band ratios and slope, newly introduced topographically-guided texture measures were found to enhance significantly the classification, and also feature selection revealed positive influence on the overall performance. With an iterative procedure to examine the class-imbalance within the training sample it was furthermore possible to compensate spurious effects of class-imbalance and class-overlap on the balance of the error rates. Employing approximately 20% of the data for training, the proposed workflow resulted in accuracies between 73% and 87% for the affected areas, and approximately balanced commission and omission errors.

A hierarchical methodology for the mesoscale assessment of building integrated roof solar energy systems

Buildings and other engineered structures that form cities are responsible for a significant portion of the global and local impacts of climate change. Consequently, the installation of building integrated renewable energy sources such as photovoltaic or solar thermal systems on building rooftops is being widely investigated. Although the advantages for individual buildings have been studied, as yet there is little understanding of the potential benefits of urban scale implementation of such systems. Here we report the development of a new methodology for assessing the potential capacity and benefits of installing rooftop photovoltaic systems in an urbanized area. Object oriented image analysis and geographical information systems are combined with remote sensing image data to quantify the rooftop area available for solar energy applications and a renewable energy computer simulation is included to predict the potential benefits of urban scale photovoltaic system implementation. The new methodology predicts energy generation potential that can be utilized to meet Arizona’s Renewable Portfolio Standard 2025 renewable energy generation requirements.

Combining Random Forests and object-oriented analysis for landslide mapping from very high resolution imagery

The increasing availability of very high resolution (VHR) remote sensing images has been leading to new opportunities for the cartography of landslides in risk management and disaster response. Object-oriented image analysis has become one of the key-concepts to better exploit additional spatial, spectral and contextual information. The multitude of additional object attributes calls for the use of advanced data mining and machine learning tools to identify the most suitable features and handle the non-linear classification task. In this study we used the Random Forest algorithm for the selection of useful features and object classification in the context of landslide mapping. A workflow for image segmentation, feature extraction, feature selection and classification was developed and tested with multi-sensor optical imagery from four different test sites. Due to class imbalance and class overlap between landslide and non-landslide areas the classifier can be heavily biased towards over- and under-prediction of the affected areas. This is a common issue for many real-world applications and a procedure to estimate a well-adjusted class ratio from the training samples was designed and tested. A number of potentially useful object metrics was evaluated and it was demonstrated that topographically guided texture measures provide significant enhancements. Employing 20% of the image objects for training accuracies between 73.3% and 87.1% were achieved at four different test sites.

Evaluation of soft possibilistic classifications with non-specificity uncertainty measures

The aim of this paper was to investigate the usefulness of non-specificity uncertainty measures to evaluate soft classifications of remote sensing images. In particular, we analysed whether these measures could be used to identify the difficulties found by the classifier and to estimate the classification accuracy. Two non-specificity uncertainty measures were considered, the non-specificity measure (NSp) and the U-uncertainty measure, and their behaviour was analysed to evaluate which is the most appropriate for this application. To overcome the fact that these two measures have different ranges, a normalized version (Un) of the U-uncertainty measure was used. Both measures were applied to evaluate the uncertainty of a soft classification of a very high spatial resolution multispectral satellite image, performed with an object-oriented image analysis based on a fuzzy classification. The classification accuracy was evaluated using an error matrix and the user's and producer's accuracies were computed. Two uncertainty indexes are proposed for each measure, and the correlation between the information given by them and the user's and producer's accuracies was determined to assess the relationship and compatibility of both sources of information. The results show that there is a positive correlation between the information given by the uncertainty and accuracy indexes, but mainly between the uncertainty indexes and the user's accuracy, where the correlation achieved 77%. This study shows that uncertainty indexes may be used, along with the possibility distributions, as indicators of the classification performance, and may therefore be very useful tools.

A Study of the Segmentation Scale of High-resolution Remotely Sensed Data in Chengdu Plain

In classifying the high-resolution remotely sensed data,the method applying the object-oriented image analysis is better than that applying the traditional pixel-oriented analysis.The first key step is the image segmentation for applying the object-oriented analysis.The results of the segmentation are the acquisition of a series of objects related to the real objects.The veracity of the segmentation is related to the selected segmentation scale.This paper has studied the selection of the segmentation scale of the high-resolution remotely sensed data obtained in Chengdu plain,applied different scales for image segmentation and compared the results.It is found that the best scale is 30,the same as the scale with which the mean brightness of the image objects has the maximum standard deviation.

Urban land cover and land use classification of an informal settlement area using the open-source knowledge-based system InterIMAGE

This study uses the InterIMAGE system and imagery from the QuickBird sensor for land cover and land use classification at two test sites with informal settlements in the metropolis of São Paulo, Brazil. InterIMAGE is an open source and free access system for knowledge-based image classification. Within InterIMAGE human knowledge is represented as a semantic net and by user-defined rules based on the paradigms of object-oriented image analysis. In the land cover classification step, a genetic algorithm was used for determining appropriate segmentation parameters. For the description of the land cover classes in terms of features and thresholds, a strategy combining machine learning algorithms and a semantic net was elaborated. Based on the land cover classifications, the land use classifications were carried out considering the urban blocks of the test sites as the analysis units. Customized features related to the composition and geometrical structures of the land cover objects within these blocks were used for the description of the land use classes. The proposed methodology has been shown to be efficient for the automatic mapping of the land cover and land use in complex urban areas. The land cover classifications achieved overall accuracies above 70 percent and Kappa indexes above 0.65. Referring to the land use classifications, overall accuracies above 87 percent and Kappa indexes above 0.71 were obtained. This study has explored the main functionalities of the InterIMAGE system, presenting its potential for object-based and knowledge-based image classification.

Delineation of Urban Footprints From TerraSAR-X Data by Analyzing Speckle Characteristics and Intensity Information

With a spatial resolution of up to 1 m, the German radar satellite TerraSAR-X (TSX) has significantly increased the usability of spaceborne synthetic aperture radar (SAR) imagery in the context of urban applications. This paper presents an approach toward the semiautomated detection of built-up areas (BAs) based on single-polarized TSX images. The proposed methodology includes a specific preprocessing of the SAR data and an automated image analysis procedure. The preprocessing focuses on the analysis of local speckle characteristics in order to provide a texture layer that highlights BAs. In the context of an object-oriented image analysis, this texture layer is used along with the original intensity information to automatically extract settlements. The technique is tested on the basis of 12 TSX scenes covering representative urban agglomerations distributed throughout the world. Overall, accuracies between 76% and 96% for the derived city footprints demonstrate the high potential of both the TSX imagery and the proposed analysis approach in detecting BAs. In order to demonstrate the robustness and transferability of the image analysis concept, we finally transferred the classification strategy from the object-oriented domain to a more general and simplified pixel-based approach.

Automated quantification and sizing of unbranched filamentous cyanobacteria by model-based object-oriented image analysis.

Quantification and sizing of filamentous cyanobacteria in environmental samples or cultures are time-consuming and are often performed by using manual or semiautomated microscopic analysis. Automation of conventional image analysis is difficult because filaments may exhibit great variations in length and patchy autofluorescence. Moreover, individual filaments frequently cross each other in microscopic preparations, as deduced by modeling. This paper describes a novel approach based on object-oriented image analysis to simultaneously determine (i) filament number, (ii) individual filament lengths, and (iii) the cumulative filament length of unbranched cyanobacterial morphotypes in fluorescent microscope images in a fully automated high-throughput manner. Special emphasis was placed on correct detection of overlapping objects by image analysis and on appropriate coverage of filament length distribution by using large composite images. The method was validated with a data set for Planktothrix rubescens from field samples and was compared with manual filament tracing, the line intercept method, and the Utermöhl counting approach. The computer program described allows batch processing of large images from any appropriate source and annotation of detected filaments. It requires no user interaction, is available free, and thus might be a useful tool for basic research and drinking water quality control.

An Evaluation of Object-Oriented Image Analysis Techniques to Identify Motorized Vehicle Effects in Semi-arid to Arid Ecosystems of the American West

Human disturbance is a leading ecosystem stressor. Human-induced modifications include transportation networks, areal disturbances due to resource extraction, and recreation activities. High-resolution imagery and object-oriented classification rather than pixel-based techniques have successfully identified roads, buildings, and other anthropogenic features. Three commercial, automated feature-extraction software packages (Visual Learning Systems' Feature Analyst, ENVI Feature Extraction, and Definiens Developer) were evaluated by comparing their ability to effectively detect the disturbed surface patterns from motorized vehicle traffic. Each package achieved overall accuracies in the 70% range, demonstrating the potential to map the surface patterns. The Definiens classification was more consistent and statistically valid.

Object-orientated image analysis for the semi-automatic detection of dead trees following a spruce bark beetle (Ips typographus) outbreak

The survey and continuing inventory in the Bavarian Forest National Park of deadwood areas resulting from a spruce bark beetle calamity are being performed by means of visual evaluation of colour infrared aerial photographs. With the aid of the object-oriented image analysis software eCognition, it was possible to develop a partially automated method for this purpose. In order to verify the classification results, a test area was classified, and the results compared with those obtained by the previously used method. In addition, the classification results from two consecutive years were compared, and accuracy assessment methods were used to scrutinize the results. Classification in the deadwood areas yielded a total classification accuracy of 91.5%. In regard to objectivity and degree of detail, the newly developed method is significantly superior to the former method, which is based on visual interpretation with a stereo workstation. One problem, however, was the insufficient spatial accuracy of the respective orthophotos. Because of this, it was not possible to detect changes over the course of specified time intervals. Therefore, a practical application of this method would require that the orthophotos from various dates or times be precisely spatially assigned. This requirement can be achieved with the production of new orthophotos.

An object-oriented approach to assessing changes in tree cover in the Colorado Front Range 1938–1999

In the wake of numerous catastrophic wildfires, forest management policies have been implemented in recent years in the United States with the goals of reducing fire risk, including the National Fire Plan and the Healthy Forest Restoration Act. A key premise underlying these policies is that fire suppression has resulted in denser forests than were present historically that now have higher fire risk. To evaluate this premise for the northern Front Range, Colorado, we used object-oriented image analysis to compare change in tree cover delineated from historical and modern imagery. Historical photographs from 1938 and 1940 were scanned, orthorectified, and overlaid on Digital Orthoimagry Quarter Quadrangles (DOQQs) from 1999. Using an object-oriented image analysis technique, the photos were then finely segmented and classified into two classes: tree and non-tree. Trees are heterogeneous in appearance in black and white aerial photography, so we employed separate membership functions to identify four visually distinct types: ‘interior forest’, ‘isolated trees’, ‘dark forest’, and ‘edge forest’. Our classification strategy employed spatial relationships between objects in addition to spectral information, so that our classification is fairly robust to variations in illumination. Based on the classification of fine objects, we then calculated the percent tree cover within a larger set of objects for the two time periods. We estimate that average tree density across the study area increased minimally (4%) during the 60-year period, with considerable spatial variation across the landscape. The results of the analysis illustrate that, consistent with independent tree-ring evidence, the highest increase in tree density is in areas characterized by low initial density, south-facing slopes, low elevations, and ponderosa pine dominance. In contrast, the highest elevation areas dominated by mixed conifer and lodgepole pine forests revealed no significant change in tree cover. Furthermore, there is no significant difference between objects dominated by low, medium, and high departure from historical conditions, as classified in the Landfire Fire Regime Condition Class (FRCC) data product. The results of the study can help managers prioritize forest treatments aimed at restoring pre-suppression forest structure.

Systems pathology

Diagnostic tumor pathology in the context of personalized medicine has progressed from an interpretive, subjective science to a more objective, evidence-based practice. This has resulted in the development of several tissue-based, molecular-driven tests that provide information regarding prognosis and response to therapy. The challenge, however, for both the pathologist and the treating physician is how best to effectively integrate this data into a comprehensive treatment plan that includes a patient-specific risk assessment. To address this need, the authors developed a systems pathology approach to the practice of clinical molecular medicine through technical advances in object-oriented image analysis, and phenotyping at the microanatomical level using deparaffinized tissue section and quantitative biomarker multiplexing. With support vector regression for censored data, they have been able to integrate complex information and provide a patient-specific risk profile based on the clinical endpoint under investigation. Cancer 2009;115(13 suppl):3078-84. (c) 2009 American Cancer Society.

Quantification of shelterbelt characteristics using high-resolution imagery

The Agriculture and Agri-Food Canada Prairie Shelterbelt Program has distributed shelterbelt trees across the Prairie Provinces since 1903 to reduce wind erosion and other environmental benefits such as sequestering carbon and providing habitat for biodiversity. To assess the existence and conditions of shelterbelts on the landscape, visiting individual shelterbelts across each province is costly and time-consuming. High-resolution imagery offers a potentially quick and inexpensive method of identifying shelterbelts and deriving information about them. As resolution of imagery increases, more information can be extracted as ground features are becoming increasingly recognizable. Although shelterbelts could be analyzed across large sections of land, finer resolution comes at a greater price in required time and computing power. Shelterbelts were examined using spectral reflectance from multi-spectral bands and using shape, texture and other relational properties as determined with object-oriented image analysis. Principal components analysis and multiple discriminate analysis were used to identify shelterbelt characters by species. In a selected region, 93 of 97 field shelterbelts (95.8%) were correctly identified from 1:40,000 orthophotos. Spectral reflectance, variance and shape parameters were combined to differentiate among six shelterbelt species compositions.

An Examination of the Effects of Spatial Resolution and Image Analysis Technique on Indirect Fuel Mapping

The high number of forest fires occurring every year constitutes one of the major degradation factors of ecosystems, especially in the Mediterranean region. The ability of satellite sensors to cover wide areas makes them valuable tools for the prevention, detection and mapping of wildfires and fire related properties of the ecosystems. However, increased spatial resolution, coupled with a lower number of available spectral channels, makes image analysis difficult. Thus, users need to employ additional image analysis techniques in order to achieve their objectives. The aim of this work was to assess to what extent potential fuels can be discriminated within different image analysis techniques and satellite imagery products using the indirect mapping method. The main hypothesis was that very high-resolution data, coupled with object-oriented image analysis (OOIA), would improve the ability to differentiate among potential fuels. To test this hypothesis, fuel types were extracted from high and very high spatial resolution satellite imagery using object-oriented and pixel-based image analyses, respectively. The general conclusion of this study was that the use of object-oriented image analysis not only produces more accurate results, but also allows differentiation among the largest number of potential fuels. Object-oriented analysis allows better discrimination of potential fuels using very high-resolution imagery and the indirect mapping method. Six different potential fuels were classified using the object-oriented approach compared with only four when using pixel-based image analysis. The overall accuracy achieved reached approximately 80%.

Urban social vulnerability assessment with physical proxies and spatial metrics derived from air- and spaceborne imagery and GIS data

Risk management in urban planning is of increasing importance to mitigate the growing amount of damage and the increasing number of casualties caused by natural disasters. Risk assessment to support management requires knowledge about present and future hazards, elements at risk and different types of vulnerability. This article deals with the assessment of social vulnerability (SV). In the past this has frequently been neglected due to lack of data and assessment difficulties. Existing approaches for SV assessment, primarily based on community-based methods or on census data, have limited efficiency and transferability. In this article a new method based on contextual analysis of image and GIS data is presented. An approach based on proxy variables that were derived from high-resolution optical and laser scanning data was applied, in combination with elevation information and existing hazard data. Object-oriented image analysis was applied for the definition and estimation of those variables, focusing on SV indicators with physical characteristics. A reference Social Vulnerability Index (SVI) was created from census data available for the study area on a neighbourhood level and tested for parts of Tegucigalpa, Honduras. For the evaluation of the proxy-variables, a stepwise regression model to select the best explanatory variables for changes in the SVI was applied. Eight out of 47 variables explained almost 60% of the variance, whereby the slope position and the proportion of built-up area in a neighbourhood were found to be the most valuable proxies. This work shows that contextual segmentation-based analysis of geospatial data can substantially aid in SV assessment and, when combined with field-based information, leads to optimization in terms of assessment frequency and cost.

Quantification and site-specification of the support practice factor when mapping soil erosion risk associated with olive plantations in the Mediterranean island of Crete

Due to inappropriate agricultural management practices, soil erosion is becoming one of the most dangerous forms of soil degradation in many olive farming areas in the Mediterranean region, leading to significant decrease of soil fertility and yield. In order to prevent further soil degradation, proper measures are necessary to be locally implemented. In this perspective, an increase in the spatial accuracy of remote sensing datasets and advanced image analysis are significant tools necessary and efficient for mapping soil erosion risk on a fine scale. In this study, the Revised Universal Soil Loss Equation (RUSLE) was implemented in the spatial domain using GIS, while a very high resolution satellite image, namely a QuickBird image, was used for deriving cover management (C) and support practice (P) factors, in order to map the risk of soil erosion in Kolymvari, a typical olive farming area in the island of Crete, Greece. The results comprised a risk map of soil erosion when P factor was taken uniform (conventional approach) and a risk map when P factor was quantified site-specifically using object-oriented image analysis. The results showed that the QuickBird image was necessary in order to achieve site-specificity of the P factor and therefore to support fine scale mapping of soil erosion risk in an olive cultivation area, such as the one of Kolymvari in Crete. Increasing the accuracy of the QB image classification will further improve the resulted soil erosion mapping.

Historical analysis of landscape change using remote sensing techniques: An explanatory tool for agricultural transformation in Greek rural areas

The role of agricultural change in transforming Greek rural landscapes is analysed within the broader context and theory of agricultural transition from traditional to productivist and post-productivist regimes. The analysis is based on remote sensing data following a case study approach in two study sites – Tymfi and Konitsa – located in the Epirus mountain area. Vegetation maps were created for 1945, 1969 and 1995 via ‘object oriented image analysis’ of a series of air-photographs. A number of landscape metrics was then calculated for each map and the interpreted spatial analysis illustrates that the observed landscape changes are linked to agricultural restructure. During the post war period when agriculture in Western Europe had entered a productivist phase, agricultural change was marked by trends of agricultural abandonment predominantly evident in the upland areas. In contrast, farming persisted at the lowlands and the gradual aggregation of crop fields in few large patches, indicates the existence of productivist agriculture. This landscape transformation suggests that productivist agriculture and land abandonment can coexist in a spatially differentiated pattern. Although since the mid-1980s, the rationale of agricultural policy priorities shifted towards post-productivism, the findings of this study do not support the existence of such a transition. The results are further discussed within the broader context of agricultural change in Europe providing an additional dimension to the discussions regarding the spatio-temporal properties of agricultural transition. Understanding the process of agricultural change could enhance the role of agricultural policy as a tool for landscape management and regional planning.

The Importance of Scale in Object-based Mapping of Vegetation Parameters with Hyperspectral Imagery

In recent years, object-oriented image analysis has been widely adopted by the remote sensing community. Much attention has been given to its application, while the fundamental issue of scale, here characterized by spatial object-definition, seems largely neglected. In the case of vegetation parameters like aboveground biomass and leaf area index (LAI), fundamental objects are individual trees or shrubs, each of which has a specific value. Their spatial extent, however, does not match pixels in size and shape, nor does it fit the requirements of regional studies. Estimation of vegetation parameters consequently demands larger observation units, like vegetation patches, which are better represented by variably shaped objects than by square pixels. This study aims to investigate optimal object definition for biomass and LAI. We have data from 243 field plots in our test site in southern France. They cover a vegetation range from landes to garrigue to maquis, which is considered to be the climax vegetation in the area. A HyMap image covers the area. The image is subjected to a Minimum Noise Fraction (MNF) transformation, after which it is segmented with ten different heterogeneities. The result is ten object sets, each having a different mean object size. These object sets are combined with the original image with the mean band values serving as object attributes. Field observations are linked to the corresponding objects for each object set. Using Ridge regression, relations between field observations and spectral values are identified. The prediction error is determined for each object set by cross validation. The overall lowest prediction error indicates the optimal heterogeneity for segmentation. Results show that the scale of prediction affects prediction accuracy, that increasing the object size yields an optimum in prediction accuracy, and that aboveground biomass and LAI can be associated with different optimal object sizes. Furthermore, it is shown that the accuracy of parameter estimation is higher for object-oriented analysis than for per-pixel analysis.

An empirical InSAR-optical fusion approach to mapping vegetation canopy height

Exploiting synergies afforded by a host of recently available national-scale data sets derived from interferometric synthetic aperture radar (InSAR) and passive optical remote sensing, this paper describes the development of a novel empirical approach for the provision of regional- to continental-scale estimates of vegetation canopy height. Supported by data from the 2000 Shuttle Radar Topography Mission (SRTM), the National Elevation Dataset (NED), the LANDFIRE project, and the National Land Cover Database (NLCD) 2001, this paper describes a data fusion and modeling strategy for developing the first-ever high-resolution map of canopy height for the conterminous U.S. The approach was tested as part of a prototype study spanning some 62,000 km 2 in central Utah (NLCD mapping zone 16). A mapping strategy based on object-oriented image analysis and tree-based regression techniques is employed. Empirical model development is driven by a database of height metrics obtained from an extensive field plot network administered by the USDA Forest Service–Forest Inventory and Analysis (FIA) program. Based on data from 508 FIA field plots, an average absolute height error of 2.1 m ( r = 0.88) was achieved for the prototype mapping zone.

Archaeological remote sensing application pre-post war situation of Babylon archaeological site—Iraq

The first basic step in obtaining a correct geographical knowledge and initiative for archaeological cartography analysis is an adequately geo-localized representation of natural and semi-natural resources and human activities, present and past. In this context, the correct and contextual evaluation of the resources through the use of integrated techniques of aerial photos, remote sensing and geographic information system (GIS) supply the synoptic instrument to the real knowledge of the land geography and for the operational management of any research and project. We will describe, at a synthetic level, the maturity of the land systematic study of Babylon archaeological site using different change detection analysis. Topographic maps of 1920 and 1980 were used, 18 aerial photos (1986) were mosaicked and georeferenced, vector information was digitized and inserted in a GIS system, DTM was build. Object oriented image analysis activity is being carried on and initial results are available through a WebGIS. The use of remote sensing (Quickbird and Ikonos) data allows us to capture the integral mutations due to human interventions. Earth observation data and GIS system were an optimal starting point for generating and updating the cartography. This results will be indispensable for the Iraqi authority and scientific community who care about the future of the territory.