Thomas Blaschke

A multi-scale segmentation/object relationship modelling methodology for landscape analysis

Natural complexity can best be explored using spatial analysis tools based on concepts of landscape as process continuums that can be partially decomposed into objects or patches. We introduce a five-step methodology based on multi-scale segmentation and object relationship modelling. Hierarchical patch dynamics (HPD) is adopted as the theoretical framework to address issues of heterogeneity, scale, connectivity and quasi-equilibriums in landscapes. Remote sensing has emerged as the most useful data source for characterizing land use/land cover but a vast majority of applications rely on basic image processing concepts developed in the 1970s: one spatial scale, per-pixel classification of a multi-scale spectral feature space. We argue that this methodology does not make sufficient use of spatial concepts of neighbourhood, proximity or homogeneity. In contrast, the authors demonstrate in this article the utility of the HPD framework as a theoretical basis for landscape analysis in two different projects using alternative image processing methodologies, which try to overcome the ‘pixel-centred’ view. The first project focuses on habitat mapping using a high dimension multi-scale GIS database. Focal patches are derived through aggregating automatically generated landscape segments using sub-patch information including dominant tree crown densities and species. The second project uses fractal-based segmentation to produce multiple candidate segmented agricultural scenes, and then develops a decision framework to choose the combination of segmentation levels best suited to identifying shrub encroachment. The challenge and flexibility of the multi-scale segmentation/object relationship modellingapproach lies in the defining of the semantic rules which relate the lower level landscape units or holons to higher levels of organization. We seek to embrace the challenges of scale and hierarchy in landscapes and have tested two different ways to decompose complex natural environments into focal units utilising topological relations to model between the smallest units of differentiation and the focal level. We believe the use of a HPD theoretical framework will help development of better tools for characterizing the patterns and processes, acting through a range of scales, which make up landscapes.

A comparison of three image-object methods for the multiscale analysis of landscape structure

Within the conceptual framework of Complex Systems, we discuss the importance and challenges in extracting and linking multiscale objects from high-resolution remote sensing imagery to improve the monitoring, modeling and management of complex landscapes. In particular, we emphasize that remote sensing data are a particular case of the modifiable areal unit problem (MAUP) and describe how image-objects provide a way to reduce this problem. We then hypothesize that multiscale analysis should be guided by the intrinsic scale of the dominant landscape objects composing a scene and describe three different multiscale image-processing techniques with the potential to achieve this. Each of these techniques, i.e., Fractal Net Evolution Approach (FNEA), Linear Scale-Space and Blob-Feature Detection (SS), and Multiscale Object-Specific Analysis (MOSA), facilitates the multiscale pattern analysis, exploration and hierarchical linking of image-objects based on methods that derive spatially explicit multiscale contextual information from a single resolution of remote sensing imagery. We then outline the weaknesses and strengths of each technique and provide strategies for their improvement. D 2003 Elsevier Science B.V. All rights reserved.

Land cover change assessment using decision trees, support vector machines and maximum likelihood classification algorithms

Abstract Land cover change assessment is one of the main applications of remote sensed data. A number of pixel based classification algorithms have been developed over the past years for the analysis of remotely sensed data. The most notable include the maximum likelihood classifier (MLC), support vector machines (SVMs) and the decision trees (DTs). The DTs in particular offer advantages not provided by other approaches. They are computationally fast and make no statistical assumptions regarding the distribution of data. The challenge to using DTs lies in the determination of the “best” tree structure and the decision boundaries. Recent developments in the field of data mining have however, provided an alternative for overcoming the above shortcomings. In this study, we analysed the potential of DTs as one technique for data mining for the analysis of the 1986 and 2001 Landsat TM and ETM+ datasets, respectively. The results were compared with those obtained using SVMs, and MLC. Overall, acceptable accuracies of over 85% were obtained in all the cases. In general, the DTs performed better than both MLC and SVMs.

Image Segmentation Methods for Object-based Analysis and Classification

Th e continuously improving spatial resolution of remote sensing (RS) sensors sets new demand for applications utilising this information. Th e need for the more effi cient extraction of information from high resolution RS imagery and the seamless integration of this information into Geographic Information System (GIS) databases is driving geo-information theory and methodology into new territory. As the dimension of the ground instantaneous fi eld of view (GIFOV), or pixel (picture element) size, decreases many more fi ne landscape features can be readily delineated, at least visually. Th e challenge has been to produce proven man-machine methods that externalize and improve on human interpretation skills. Some of the most promising results in this research programme have come from the adoption of image segmentation algorithms and the development of so-called object-based classifi cation methodologies. In this chapter we describe diff erent approaches to image segmentation and explore how segmentation and object-based methods improve on traditional pixel-based image analysis/classifi cation methods. According to Schowengerdt () the traditional image processing/image classifi cation methodology is referred to as an image-centred approach. Here, the primary goal is to produce a map describing the spatial relationships between phenomena of interest. A second type, the data-centred approach, is pursued when the user is primarily interested in estimating parameters for individual phenomena based on the data values. Due to recent developments in image processing the two approaches appear to be converging: from image and data centred views to an information-centred approach. For instance, for change detection and environmental monitoring tasks we must not only extract information from the spectral and temporal data dimensions. We must also integrate these estimates into a spatial framework and make a priori and a posteriori utilization of GIS databases. A decision support system must encapsulate manager knowledge, context/ecological knowledge and planning knowledge. Technically, this necessitates a closer integration of remote sensing and GIS methods. Ontologically, it demands a new methodology that can provide a fl exible, demanddriven generation of information and, consequently, hierarchically structured semantic rules describing the relationships between the diff erent levels of spatial entities. Several of the aspects of geo-information involved cannot be obtained by pixel information as such but can only be achieved with an exploitation of neighbourhood information and context of the objects of interest. Th e relationship between ground objects and image objects

GIS-multicriteria decision analysis for landslide susceptibility mapping: comparing three methods for the Urmia lake basin, Iran

The GIS-multicriteria decision analysis (GIS-MCDA) technique is increasingly used for landslide hazard mapping and zonation. It enables the integration of different data layers with different levels of uncertainty. In this study, three different GIS-MCDA methods were applied to landslide susceptibility mapping for the Urmia lake basin in northwest Iran. Nine landslide causal factors were used, whereby parameters were extracted from an associated spatial database. These factors were evaluated, and then, the respective factor weight and class weight were assigned to each of the associated factors. The landslide susceptibility maps were produced based on weighted overly techniques including analytic hierarchy process (AHP), weighted linear combination (WLC) and ordered weighted average (OWA). An existing inventory of known landslides within the case study area was compared with the resulting susceptibility maps. Respectively, Dempster-Shafer Theory was used to carry out uncertainty analysis of GIS-MCDA results. Result of research indicated the AHP performed best in the landslide susceptibility mapping closely followed by the OWA method while the WLC method delivered significantly poorer results. The resulting figures are generally very high for this area, but it could be proved that the choice of method significantly influences the results.

Land suitability analysis for Tabriz County, Iran: a multi-criteria evaluation approach using GIS

In our research we investigated the optimal utilization of land resources for agricultural production in Tabriz County, Iran. A GIS-based Multi Criteria Decision Making land suitability analysis was performed. Hereby, several suitability factors including soils, climatic conditions, and water availability were evaluated, based on expert knowledge from stakeholders at various levels. An Analytical Hierarchical Process was used to rank the various suitability factors and the resulting weights were used to construct the suitability map layers. In doing so, the derived weights were used, and subsequently land suitability maps for irrigated and dry-farm agriculture were created. Finally, a synthesized land suitability map was generated by combining these maps and by comparing the product with current land use SPOT 5 satellite images. The resulting suitability maps indicate the areas, in which the intensity of land use for agriculture should increase, decrease or remain unchanged. Our investigations have revealed that 65676 hectares may be suitable for irrigation and 120872 hectares may be suitable for dry-farm agriculture. This indicates a substantial potential to satisfy the significantly increasing regional demand for agricultural products. The results of our research have been provided to the regional authorities and will be used in strategic land use planning.

Detecting informal settlements from QuickBird data in Rio de Janeiro using an object based approach

Informal settlements behave very dynamical over space and time and the number of people living in such housing areas is growing worldwide. The reasons for this dynamical behavior are manifold and are not matter of this article. Nevertheless, informal settlements represent a status quo of housing and living conditions which is from a humanitarian point of view in the most cases below acceptable levels. Therefore, reliable spatial information about informal settlements is vital for any actions of improvement of these living conditions. Since remote sensing data is a well suited data source for mapping and monitoring we demonstrate a methodology to detect informal settlements (favelas) from QuickBird data using an object based approach. On the one hand we therefore use experiences and adapt them which were already made by Hofmann, P. (2001) regarding the image segmentation of an IKONOS scene of Cape Town. On the other hand we resort to a general ontology of informal settlements which we then transfer to a fuzzy-logic rule base which acts as basic classifier of the generated segments. This basic rule base is than extended in a way that features of segregation given by the ontology (namely neighborhood) are applied to the extraction method as an iterative process (i.e. a knowledge based region growing). Finally, we assess the results of the simple and iterative method by comparing them with the results of a manual mapping.

Object-based contextual image classification built on image segmentation

The continuously improving spatial resolution of remote sensing sensors sets new demand for applications utilizing this information. The need for the more efficient extraction of information from high resolution RS imagery and the seamless integration of this information into Geographic Information System (GIS) databases is driving geo-information theory, and methodology, into new territory. As the dimension of the ground instantaneous field of view (GIFOV), or pixel size, decreases many more fine landscape features can be readily delineated, at least visually. The challenge has been to produce proven man-machine methods that externalize and improve on human interpretation skills. Some of the most promising results come from the adoption of image segmentation algorithms and the development of so-called object-based classification methodologies. This paper builds on a discussion of different approaches to image segmentation techniques and demonstrates through several applications how segmentation and object-based methods improve on pixel-based image analysis/classification methods. In contrast to pixel-based procedure, image objects can carry many more attributes than only spectral information. In this paper, I address the concepts of object-based image processing, and present an approach that integrates the concept of object-based processing into the image classification process. Object-based processing not only considers contextual information but also information about the shape of and the spatial relations between the image regions.

Collective Sensing: Integrating Geospatial Technologies to Understand Urban Systems—An Overview

Cities are complex systems composed of numerous interacting components that evolve over multiple spatio-temporal scales. Consequently, no single data source is sufficient to satisfy the information needs required to map, monitor, model, and ultimately understand and manage our interaction within such urban systems. Remote sensing technology provides a key data source for mapping such environments, but is not sufficient for fully understanding them. In this article we provide a condensed urban perspective of critical geospatial technologies and techniques: (i) Remote Sensing; (ii) Geographic Information Systems; (iii) object-based image analysis; and (iv) sensor webs, and recommend a holistic integration of these technologies within the language of open geospatial consortium (OGC) standards in-order to more fully understand urban systems. We then discuss the potential of this integration and conclude that this extends the monitoring and mapping options beyond “hard infrastructure” by addressing “humans as sensors”, mobility and human-environment interactions, and future improvements to quality of life and of social infrastructures.

Ontology-Based Classification of Building Types Detected from Airborne Laser Scanning Data

Abstract: Accurate information on urban building types plays a crucial role for urban development, planning, and management. In this paper, we apply Object-Based Image Analysis (OBIA) methods to extract buildings from Airborne Laser Scanner (ALS) data and investigate the possibility of classifying detected buildings into ―Residential/Small Buildings‖, ―Apartment Buildings‖, and ―Industrial and Factory Building‖ classes by means of domain ontology and machine learning techniques. The buildings objects are classified using exclusively the information computed from the ALS data. To select the relevant features for predicting the classes of interest, the Random Forest classifier has been applied. The ontology-based classification yielded convincing results for the ―Residential/Small Buildings‖ class (F-Measure 97.7%), whereas the ―Apartment Buildings‖ and ―Industrial and Factory Buildings‖ classes achieved less accurate results (F-Measure 60% and 51%, respectively).