Moritz Lennert

An Open-Source Semi-Automated Processing Chain for Urban Object-Based Classification

This study presents the development of a semi-automated processing chain for urban object-based land-cover and land-use classification. The processing chain is implemented in Python and relies on existing open-source software GRASS GIS and R. The complete tool chain is available in open access and is adaptable to specific user needs. For automation purposes, we developed two GRASS GIS add-ons enabling users (1) to optimize segmentation parameters in an unsupervised manner and (2) to classify remote sensing data using several individual machine learning classifiers or their prediction combinations through voting-schemes. We tested the performance of the processing chain using sub-metric multispectral and height data on two very different urban environments: Ouagadougou, Burkina Faso in sub-Saharan Africa and Liege, Belgium in Western Europe. Using a hierarchical classification scheme, the overall accuracy reached 93% at the first level (5 classes) and about 80% at the second level (11 and 9 classes, respectively).

Less is more: Optimizing classification performance through feature selection in a very-high-resolution remote sensing object-based urban application

This study evaluates the impact of four feature selection (FS) algorithms in an object-based image analysis framework for very-high-resolution land use-land cover classification. The selected FS algorithms, correlation-based feature selection, mean decrease in accuracy, random forest (RF) based recursive feature elimination, and variable selection using random forest, were tested on the extreme gradient boosting, support vector machine, K-nearest neighbor, RF, and recursive partitioningclassifiers, respectively. The results demonstrate that the selection of an appropriate FS method can be crucial to the performance of a machine learning classifier in terms of accuracy but also parsimony. In this scope, we propose a new metric to perform model selection named classification optimization score (COS) that rewards model simplicity and indirectly penalizes for increased computational time and processing requirements using the number of features for a given classification model as a surrogate. Our findings suggest that applying rigorous FS along with utilizing the COS metric may significantly reduce the processing time and the storage space while at the same time producing higher classification accuracy than using the initial dataset.

A local segmentation parameter optimization approach for mapping heterogeneous urban environments using VHR imagery

Mapping large heterogeneous urban areas using object-based image analysis (OBIA) remains challenging, especially with respect to the segmentation process. This could be explained both by the complex arrangement of heterogeneous land-cover classes and by the high diversity of urban patterns which can be encountered throughout the scene. In this context, using a single segmentation parameter to obtain satisfying segmentation results for the whole scene can be impossible. Nonetheless, it is possible to subdivide the whole city into smaller local zones, rather homogeneous according to their urban pattern. These zones can then be used to optimize the segmentation parameter locally, instead of using the whole image or a single representative spatial subset. This paper assesses the contribution of a local approach for the optimization of segmentation parameter compared to a global approach. Ouagadougou, located in sub-Saharan Africa, is used as case studies. First, the whole scene is segmented using a single globally optimized segmentation parameter. Second, the city is subdivided into 283 local zones, homogeneous in terms of building size and building density. Each local zone is then segmented using a locally optimized segmentation parameter. Unsupervised segmentation parameter optimization (USPO), relying on an optimization function which tends to maximize both intra-object homogeneity and inter-object heterogeneity, is used to select the segmentation parameter automatically for both approaches. Finally, a land-use/land-cover classification is performed using the Random Forest (RF) classifier. The results reveal that the local approach outperforms the global one, especially by limiting confusions between buildings and their bare-soil neighbors.

Normalization in Unsupervised Segmentation Parameter Optimization: A Solution Based on Local Regression Trend Analysis

In object-based image analysis (OBIA), the appropriate parametrization of segmentation algorithms is crucial for obtaining satisfactory image classification results. One of the ways this can be done is by unsupervised segmentation parameter optimization (USPO). A popular USPO method does this through the optimization of a “global score” (GS), which minimizes intrasegment heterogeneity and maximizes intersegment heterogeneity. However, the calculated GS values are sensitive to the minimum and maximum ranges of the candidate segmentations. Previous research proposed the use of fixed minimum/maximum threshold values for the intrasegment/intersegment heterogeneity measures to deal with the sensitivity of user-defined ranges, but the performance of this approach has not been investigated in detail. In the context of a remote sensing very-high-resolution urban application, we show the limitations of the fixed threshold approach, both in a theoretical and applied manner, and instead propose a novel solution to identify the range of candidate segmentations using local regression trend analysis. We found that the proposed approach showed significant improvements over the use of fixed minimum/maximum values, is less subjective than user-defined threshold values and, thus, can be of merit for a fully automated procedure and big data applications.

An open-source semi-automated processing chain for urban obia classification

This study presents the development of a semi-automated processing chain for OBIA urban land-cover and land-use classification. Implemented in Python and relying on existing open-source software GRASS GIS and R. The complete tool chain is available in open-access and adaptable to specific user needs. For automation purpose, we developed two GRASS GIS add-ons allowing (1) to optimize segmentation parameters in an unsupervised manner and (2) to classify remote sensing data using several individual machine learning classifiers or their predictions combination through voting-schemes. We tested the performance and transferability of the processing chain using sub-metric multispectral and height data on two very different urban environments: Ouagadougou, Burkina Faso in sub-Saharan Africa and Liege, Belgium in Western Europe. Using a hierarchical classification scheme, the kappa values reached for both cities about 0.78 at the second level (9 and 11 classes) and 0.90 at the first level (5 classes).

Toward an operational framework for fine-scale urban land-cover mapping in Wallonia using submeter remote sensing and ancillary vector data

Abstract. Encouraged by the EU INSPIRE directive requirements and recommendations, the Walloon authorities, similar to other EU regional or national authorities, want to develop operational land-cover (LC) and land-use (LU) mapping methods using existing geodata. Urban planners and environmental monitoring stakeholders of Wallonia have to rely on outdated, mixed, and incomplete LC and LU information. The current reference map is 10-years old. The two object-based classification methods, i.e., a rule- and a classifier-based method, for detailed regional urban LC mapping are compared. The added value of using the different existing geospatial datasets in the process is assessed. This includes the comparison between satellite and aerial optical data in terms of mapping accuracies, visual quality of the map, costs, processing, data availability, and property rights. The combination of spectral, tridimensional, and vector data provides accuracy values close to 0.90 for mapping the LC into nine categories with a minimum mapping unit of 15  m2. Such a detailed LC map offers opportunities for fine-scale environmental and spatial planning activities. Still, the regional application poses challenges regarding automation, big data handling, and processing time, which are discussed.

Scale Matters: Spatially Partitioned Unsupervised Segmentation Parameter Optimization for Large and Heterogeneous Satellite Images

To classify Very-High-Resolution (VHR) imagery, Geographic Object Based Image Analysis (GEOBIA) is the most popular method used to produce high quality Land-Use/Land-Cover maps. A crucial step in GEOBIA is the appropriate parametrization of the segmentation algorithm prior to the classification. However, little effort has been made to automatically optimize GEOBIA algorithms in an unsupervised and spatially meaningful manner. So far, most Unsupervised Segmentation Parameter Optimization (USPO) techniques, assume spatial stationarity for the whole study area extent. This can be questionable, particularly for applications in geographically large and heterogeneous urban areas. In this study, we employed a novel framework named Spatially Partitioned Unsupervised Segmentation Parameter Optimization (SPUSPO), which optimizes segmentation parameters locally rather than globally, for the Sub-Saharan African city of Ouagadougou, Burkina Faso, using WorldView-3 imagery (607 km2). The results showed that there exists significant spatial variation in the optimal segmentation parameters suggested by USPO across the whole scene, which follows landscape patterns—mainly of the various built-up and vegetation types. The most appropriate automatic spatial partitioning method from the investigated techniques, was an edge-detection cutline algorithm, which achieved higher classification accuracy than a global optimization, better predicted built-up regions, and did not suffer from edge effects. The overall classification accuracy using SPUSPO was 90.5%, whilst the accuracy from undertaking a traditional USPO approach was 89.5%. The differences between them were statistically significant (p < 0.05) based on a McNemar’s test of similarity. Our methods were validated further by employing a segmentation goodness metric, Area Fit Index (AFI)on building objects across Ouagadougou, which suggested that a global USPO was more over-segmented than our local approach. The mean AFI values for SPUSPO and USPO were 0.28 and 0.36, respectively. Finally, the processing was carried out using the open-source software GRASS GIS, due to its efficiency in raster-based applications.

Contribution of nDSM derived from VHR stereo imagery to urban land-cover mapping in Sub-Saharan Africa

Mapping the urban land cover from VHR optical imagery remains a challenging task, more particularly in cities that present complex landscapes and patterns. In this study, we assessed the contribution of height data derived from WorldView-3 stereo imagery for mapping the land cover of Sub-Saharan African cities. Our case study is located in Ouagadougou, Burkina Faso. An OBIA approach was implemented using an open-source semi-automated processing chain. The use of the nDSM as input to the segmentation and/or to the classification in addition to the four VNIR WorldView-3 optical bands was evaluated. The quantitative and qualitative analysis of the results indicate an improvement for a number of classes, among which the class ‘Buildings’ that is of particular interest in many applications. Visually, this improvement is more noticeable in planned settlements and industrial areas than in unplanned settlements.

Optimizing classification performance in an object-based very-high-resolution land use-land cover urban application

This study evaluates the impact of three Feature Selection (FS) algorithms in an Object Based Image Analysis (OBIA) framework for Very-High-Resolution (VHR) Land Use-Land Cover (LULC) classification. The three selected FS algorithms, Correlation Based Selection (CFS), Mean Decrease in Accuracy (MDA) and Random Forest (RF) based Recursive Feature Elimination (RFE), were tested on Support Vector Machine (SVM), K-Nearest Neighbor, and Random Forest (RF) classifiers. The results demonstrate that the accuracy of SVM and KNN classifiers are the most sensitive to FS. The RF appeared to be more robust to high dimensionality, although a significant increase in accuracy was found by using the RFE method. In terms of classification accuracy, SVM performed the best using FS, followed by RF and KNN. Finally, only a small number of features is needed to achieve the highest performance using each classifier. This study emphasizes the benefits of rigorous FS for maximizing performance, as well as for minimizing model complexity and interpretation.

Building applications with FOSS4G bricks: two examples of the use of GRASS GIS modules as a high-level “language” for the analyses of continuous space data in economic geography

In a world where researchers are more and more confronted to large sets of microdata, new algorithms are constantly developed that have to be translated into usable programs. Modular GIS toolkits such as GRASS GIS offer a middle way between low-level programming approaches and GUI-based desktop GIS. The modules can be seen as elements of a programming language which makes the implementation of algorithms for spatial analysis very easy for researchers. Using two examples of algorithms in economic geography, for estimating regional exports and for determining raster-object neighbourhood matrices, this paper shows how just a few module calls can replace more complicated low-level programs, as long as the researcher can change perspective from a pixel-by-pixel view to a map view of the problem at hand. Combining GRASS GIS with Python as general glue between modules also offers options for easy multi-processing, as well as supporting the increasingly loud call for open research, including open source computing tools in research. * Corresponding author FOSS