Irmgard Niemeyer

Change detection using object features

For the detection of changes, several statistical techniques exist. When adopted to high-resolution imagery, the results of traditional pixel-based algorithms are often limited. We propose an unsupervised change detection and classification procedure based on object features. Following the automatic pre-processing of the image data, image objects and their object features are extracted. Change detection is performed by the multivariate alteration detection (MAD), accompanied by the maximum autocorrelation factor (MAF) transformation. The change objects are then classified using the fuzzy maximum likelihood estimation (FMLE). Finally the classification of changes is improved by probabilistic label relaxation.

A procedure for automatic object-based classification

A human observer can easily categorize an image into classes of interest but it is generally difficult to reproduce the same result using a computer. The emerging object-based methodology for image classification appears to be a better way to mimic the human thought process. Unlike pixel-based techniques which only use the layer pixel values, the object-based techniques can also use shape and context information of a scene texture. These extra degrees of freedom provided by the objects will aid the identification (or classification) of visible textures. However, the concept of image-objects brings with it a large number of object features and thus a lot of information is associated with the objects. In this article, we present a procedure for object-based classification which effectively utilizes the huge information associated with the objects and automatically generates classification rules. The solution of automation depends on how we solve the problem of identifying the features that characterize the classes of interest and then finding the final distribution of the classes in the identified feature space. We try to illustrate the procedure applied for a two-class case and then suggest possible ways to extend the method for multiple classes.

Feature recognition in the context of automated object-oriented analysis of remote sensing data monitoring the iranian nuclear sites

Against the background of nuclear safeguards applications using commercially available satellite imagery, procedures for wide-area monitoring of the Iranian nuclear fuel cycle are investigated. Specifically, object-oriented classification combined with statistical change detection is applied to high-resolution imagery. In this context, a feature recognition and analysis tool, called SEaTH, has been developed for automatic selection of optimal object class features for subsequent classification. The application of SEaTH is presented in a case study of the NFRPC Esfahan, Iran. The transferability of classification models is discussed regarding the necessity for automation of extensive monitoring tasks.

Automation of change detection procedures for nuclear safeguards-related monitoring purposes

Against the background of nuclear safeguards applications using commercially available satellite imagery, a two-steps attempt for change detection and analysis was realized in general. Beginning with the wide-area monitoring on the basis of medium-resolution satellite data for the pre-scanning of significant changes within the nuclear-related locations, the areas of interest could then be explicitly analyzed by change detection and analysis methods using high-resolution satellite data. The change pixels were detected by using the multivariate alteration detection (MAD) transformation, producing a set of mutually orthogonal difference images (the so-called MAD variates). The decision thresholds for the change pixels were set by applying a probability mixture model to the MAD variates based on an EM algorithm. By means of eCognition a second, object-oriented procedure was implemented in order to create an automated workflow for the multiscale extraction of the (change) objects and (change) features for the subsequent post-classification of the areas of interest. Regarding the necessity of automation for extensive monitoring tasks the processing aspects of standardization and transferability took the centre stage of the investigations.

Hyperspectral data classification using an ensemble of class-dependent neural networks

Hyperspectral data are characterized by a huge size due to hundreds of narrow frequency bands. However, the classes of interest are often characterized by only a few features from the available (ormodified) feature space. Using a few samples of the classes of interest it is possible to identify the features characterizing the classes by calculating the Bhattacharya distance, B or the Jeffries-Matusita distance, J for every feature and for every class combination. However, the classification using these features is not trivial. We use a new architecture based on class-dependent neural networks for this purpose. Class-dependent neural network is a feed forward neural network for every class with features characterizing only that class as inputs. In the combined architecture, all the classes are first, individually separated from other classes using first level class-dependent neural networks to map the characteristic features to a fuzzy value for each of the classes. Then, a final classification decision is made using a second level neural network with inputs from the outputs of the first level neural networks.

Change Detection: The Potential for Nuclear Safeguards

An object-oriented monitoring system for nuclear safeguards purposes was proposed in order to detect changes within nuclear facilities. By means of pixel-based change detection and object-oriented post-classification by eCognition some investigations were carried out in terms of automation, thus standardization and transferability. As a result, medium-resolution imagery could be considered as suitably for change-/no change-analysis in terms of wide area monitoring, for the detailed object-oriented analysis of significant changes high-resolution satellite imagery should be used. The automation and the transferability of the change detection and analysis procedures appears to be feasible to a certain extent, therewith giving rough and fast indications of areas of interest and explicitly analyzing the relevant areas.

Change detection using the object features

For the detection of changes, several statistical techniques exist. When adopted to high-resolution imagery, the results of the traditional pixel-based algorithms are often limited. Especially if small structural changes are to be detected, object- based procedures show promises. In the given paper, we propose an unsupervised object-based change detection and change classification approach based on the object features. Following the automatic pre-processing, image objects and their object features are extracted. Change detection is performed by the multivariate alteration detection (MAD), accompanied by the maximum autocorrelation factor (MAF) transformation. The change objects are then classified using the fuzzy maximum likelihood estimation (FMLE). Finally the classification of changes is improved by probabilistic label relaxation.

Change Detection Tools

In this chapter a wide range of change detection tools is addressed. They are grouped into methods suitable for optical and multispectral data, synthetic aperture radar (SAR) images, and 3D data. Optical and multispectral methods include unsupervised approaches, supervised and knowledge-based approaches, pixel-based and object-oriented approaches, multivariate alteration detection, hyperspectral approaches, and approaches that deal with changes between optical images and existing vector data. Radar methods include constant false-alarm rate detection, adaptive filtering, multi-channel segmentation (an object-oriented approach), hybrid methods, and coherent change detection. 3D methods focus on tools that are able to deal with 3D information from ground based laser-ranging systems, LiDAR, and elevation models obtained from air/space borne optical and SAR data. Highlighted applications are landcover change, which is often one of the basic types of information to build analysis on, monitoring of nuclear safeguards, third-party interference close to infrastructures (or borders), and 3D analysis. What method to use is dependent on the sensor, the size of the changes in comparison with the resolution, their shape, textural properties, spectral properties, and behaviour in time, and the type of application. All these issues are discussed to be able to determine the right method, with references for further reading

Evaluation of the Efficiency of Object-Based Classification in the Identification of Geological Structures Case Study: Extraction of the Morphology of the Normal Faults

Object-based classification is a promising method- ology. Unlike pixel-based techniques which only use the layer pixel values, the object-based techniques can also use shape and context information of a scene texture. These degrees of freedom provided by the objects will aid the identification of geological structures. In this article, we present an evaluation of object- based classification in the context of extracting morphology of geological faults. An automatic classification procedure is prepared to extract the faults. The DEM and radar images of an area near Lake Magadi, Kenya, are processed separately to identify which of them is a better candidate for mapping faults. Further, after classifying the faults, it is interesting to see how the notion of an object helps in determining the statistics of the faults populations.

Detection of land cover changes in El Rawashda Forest, Sudan: A systematic comparison

This study compares six change detection techniques to study land cover change associated with tropical forest (El Rawashda forest reserve, Gedaref State, Sudan). For this site, Landsat 7 Enhanced Thematic Mapper (ETM+) data acquired on March 22, 2003 and Aster data acquired on February 26, 2006 were used. The change detection techniques employed in this study were Post-Classification Comparison (PCC), image differencing of different vegetation indices (Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI) and Transformed Difference Vegetation Index (TDVI)), Principal Component Analysis (PCA), Multivariate Alteration Detection (MAD), Change Vector Analysis (CVA) and Tasseled Cap Analysis (TCA). As field validation data did not exist for 2003, a manual classification was performed, then a change map was conducted to locate and identify change in vegetation. This change map was used as a reference to quantitatively assess the accuracy of each change-detection techniques. Based on accuracy assessment, the most successful technique was the PCC technique with an accuracy of 94%. This was followed by the MAD technique with an accuracy 88.8%. However, among vegetation indices techniques, TDVI stood out as better than NDVI and SAVI in its ability to accurately identify vegetation change.