Tessio Novack

Machine Learning Comparison between WorldView-2 and QuickBird-2-Simulated Imagery Regarding Object-Based Urban Land Cover Classification

The objective of this study is to compare WorldView-2 (WV-2) and QuickBird-2-simulated (QB-2) imagery regarding their potential for object-based urban land cover classification. Optimal segmentation parameters were automatically found for each data set and the obtained results were quantitatively compared and discussed. Four different feature selection algorithms were used in order to verify to which data set the most relevant object-based features belong to. Object-based classifications were performed with four different supervised algorithms applied to each data set and the obtained accuracies and model performances indexes were compared. Segmentation experiments carried out involving bands exclusively available in the WV-2 sensor generated segments slightly more similar to our reference segments (only about 0.23 discrepancy). Fifty seven percent of the different selected features and 53% of all the 80 selections refer to features that can only be calculated with the additional bands of the WV-2 sensor. On the other hand, 57% of the most relevant features and 63% of the second most relevant features can also be calculated considering only the QB-2 bands. In 10 out of 16 classifications, higher Kappa values were achieved when features related to the additional bands of the WV-2 sensor were also considered. In most cases, classifications carried out with the 8-band-related features generated less complex and more efficient models than those generated only with QB-2 band-related features. Our results lead to the conclusion that spectrally similar classes like ceramic tile roofs and bare soil, as well as asphalt and dark asbestos roofs can be better distinguished when the additional bands of the WV-2 sensor are used throughout the object-based classification process.

A knowledge-based, transferable approach for block-based urban land-use classification

In this work we propose a knowledge-based approach for land-use classification of city blocks through the automatic interpretation of very-high-resolution remote-sensing imagery. Our approach is founded on geographic object-based image analysis (GEOBIA) concepts and is concerned with transferability across distinct knowledge representation formalisms. This paper therefore investigates the viability of translating a high-level description of the interpretation problem into the particular knowledge representation structures and interpretation strategies of two different software platforms, namely the proprietary Definiens Developer system and the open-source InterIMAGE system. Initially, textual descriptions of the land-use classes of interest were created by photo interpreters. Then, generic class descriptions were defined as a system-independent knowledge model, which was subsequently translated into interpretation projects in the different systems. Altogether 49 blocks located on two different test-sites in the city of São Paulo (Brazil) were considered in the experiments. Although the classification results from the Definiens Developer system were slightly better than those obtained with the InterIMAGE system, we concluded that both systems have been shown to be equally qualified to implement the target application properly through adaptation of the generic knowledge model.

Classifying the Built-Up Structure of Urban Blocks with Probabilistic Graphical Models and TerraSAR-X Spotlight Imagery

Up-to-date maps of a city’s urban structure types (USTs) are very important for effective planning, as well as for different studies and applications. We present an approach for the classification of USTs at the level of urban blocks based on high-resolution spaceborne radar imagery. Images obtained at the satellite’s ascending and descending orbits were used for extracting line and polygon features from each block. Most of the attributes considered in the classification concern the geometry of these features, as well as their spatial disposition inside the blocks. Furthermore, assuming the UST classes of neighboring blocks are probabilistically dependent, we explored the framework of probabilistic graphical models and propose different context-based classification models. These models differ with respect to (i) their type, i.e., Markov or conditional random fields, (ii) their parameterization and (iii) the criterion applied for establishing pairwise neighboring relationships between blocks. In our experiments, 1,695 blocks from the city of Munich (Germany) and five representative UST classes were considered. A standard classification performed with the Random Forest algorithm achieved an overall accuracy of nearly 70%. All context-based classifications achieved overall accuracies up to 7% higher than that. The results indicate that denser pairwise block-neighborhood structures lead to better results and that the accuracy improvement is higher when the strength of the contextual influences is proportional to the similarity of the neighboring blocks attributes.

Feature selection analysis of WorldView-II data for similar urban objects distinction

Feature selection methods have recently become very attractive to remote sensing researches as new object-based image analysis systems have made available tens or even hundreds of spectral, textural and geometrical features to be used in classification routines. The aim of this study is to investigate, based on different feature selection algorithms, whether the additional bands of the WorldView II sensor are indeed helpful for the discrimination of similar urban targets.

Estimation of Population Density of Census Sectors Using Remote Sensing Data and Spatial Regression

Assuming that urban planning aims the optimization of urban functioning and the well-being of citizens, questions like “how many people are living in the city?” and “where do they live?” become key issues. In this work we utilized landscape metrics generated by the FragStats software for the estimation of population density out of census sectors in the mega city of Sao Paulo, Brazil. The metrics were calculated over an image from the QuickBird II sensor classified by the Maximum Likelihood algorithm. The accuracy of the classified image was analyzed qualitatively. Ordinary linear regression models were generated and formal statistical tests applied. The residuals from each model had its spatial dependency analyzed by visualizing its LISA Maps and by the Global Moran index. Afterwards, spatial regression models were tried and a significant improvement was obtained in terms of spatial dependency reduction and increase of the prediction power of the models. For the sake of comparison, the use of dummy variables was also tried and it became a suitable option for eliminating spatial dependency of the residuals as well. The results proved that some landscape metrics obtained over high resolution images, classified by simple supervised methods, can predict well the population density at the area under study when using it as independent variable in spatial regression models.