Ribana Roscher

Can I Trust My One-Class Classification?

Contrary to binary and multi-class classifiers, the purpose of a one-class classifier for remote sensing applications is to map only one specific land use/land cover class of interest. Training these classifiers exclusively requires reference data for the class of interest, while training data for other classes is not required. Thus, the acquisition of reference data can be significantly reduced. However, one-class classification is fraught with uncertainty and full automatization is difficult, due to the limited reference information that is available for classifier training. Thus, a user-oriented one-class classification strategy is proposed, which is based among others on the visualization and interpretation of the one-class classifier outcomes during the data processing. Careful interpretation of the diagnostic plots fosters the understanding of the classification outcome, e.g., the class separability and suitability of a particular threshold. In the absence of complete and representative validation data, which is the fact in the context of a real one-class classification application, such information is valuable for evaluation and improving the classification. The potential of the proposed strategy is demonstrated by classifying different crop types with hyperspectral data from Hyperion.

Incremental Import Vector Machines for Classifying Hyperspectral Data

In this paper, we propose an incremental learning strategy for import vector machines (IVM), which is a sparse kernel logistic regression approach. We use the procedure for the concept of self-training for sequential classification of hyperspectral data. The strategy comprises the inclusion of new training samples to increase the classification accuracy and the deletion of noninformative samples to be memory and runtime efficient. Moreover, we update the parameters in the incremental IVM model without retraining from scratch. Therefore, the incremental classifier is able to deal with large data sets. The performance of the IVM in comparison to support vector machines (SVM) is evaluated in terms of accuracy, and experiments are conducted to assess the potential of the probabilistic outputs of the IVM. Experimental results demonstrate that the IVM and SVM perform similar in terms of classification accuracy. However, the number of import vectors is significantly lower when compared to the number of support vectors, and thus, the computation time during classification can be decreased. Moreover, the probabilities provided by IVM are more reliable, when compared to the probabilistic information, derived from an SVMs output. In addition, the proposed self-training strategy can increase the classification accuracy. Overall, the IVM and its incremental version is worthwhile for the classification of hyperspectral data.

Editors Choice Article: I2VM: Incremental import vector machines

We introduce an innovative incremental learner called incremental import vector machines (I^2VM). The kernel-based discriminative approach is able to deal with complex data distributions. Additionally, the learner is sparse for an efficient training and testing and has a probabilistic output. We particularly investigate the reconstructive component of import vector machines, in order to use it for robust incremental learning. By performing incremental update steps, we are able to add and remove data samples, as well as update the current set of model parameters for incremental learning. By using various standard benchmarks, we demonstrate how I^2VM is competitive or superior to other incremental methods. It is also shown that our approach is capable of managing concept-drifts in the data distributions.

Kernel Discriminative Random Fields for land cover classification

Logistic Regression has become a commonly used classifier, not only due to its probabilistic output and its direct usage in multi-class cases. We use a sparse Kernel Logistic Regression approach - the Import Vector Machines - for land cover classification. We improve our segmentation results applying a Discriminative Random Field framework on the probabilistic classification output. We consider the performance regarding to the classification accuracy and the complexity and compare it to the Gaussian Maximum Likelihood classification and the Support Vector Machines.

Shapelet-Based Sparse Representation for Landcover Classification of Hyperspectral Images

This paper presents a sparse-representation-based classification approach with a novel dictionary construction procedure. By using the constructed dictionary, sophisticated prior knowledge about the spatial nature of the image can be integrated. The approach is based on the assumption that each image patch can be factorized into characteristic spatial patterns, also called shapelets, and patch-specific spectral information. A set of shapelets is learned in an unsupervised way, and spectral information is embodied by training samples. A combination of shapelets and spectral information is represented in an undercomplete spatial-spectral dictionary for each individual patch, where the elements of the dictionary are linearly combined to a sparse representation of the patch. The patch-based classification is obtained by means of the representation error. Experiments are conducted on three well-known hyperspectral image data sets. They illustrate that our proposed approach shows superior results in comparison to sparse-representation-based classifiers that use only limited spatial information and behaves competitively with or better than state-of-the-art classifiers utilizing spatial information and kernelized sparse-representation-based classifiers.

Superpixel-based classification of hyperspectral data using sparse representation and conditional random fields

This paper presents a superpixel-based classifier for landcover mapping of hyperspectral image data. The approach relies on the sparse representation of each pixel by a weighted linear combination of the training data. Spatial information is incorporated by using a coarse patch-based neighborhood around each pixel as well as data-adapted superpixels. The classification is done via a hierarchical conditional random field, which utilizes the sparse-representation output and models spatial and hierarchical structures in the hyperspectral image. The experiments show that the proposed approach results in superior accuracies in comparison to sparse-representation based classifiers that solely use a patch-based neighborhood.

Statistical Inference, Learning and Models in Big Data

Summary The need for new methods to deal with big data is a common theme in most scientific fields, although its definition tends to vary with the context. Statistical ideas are an essential part of this, and as a partial response, a thematic program on statistical inference, learning and models in big data was held in 2015 in Canada, under the general direction of the Canadian Statistical Sciences Institute, with major funding from, and most activities located at, the Fields Institute for Research in Mathematical Sciences. This paper gives an overview of the topics covered, describing challenges and strategies that seem common to many different areas of application and including some examples of applications to make these challenges and strategies more concrete.

Subpixel Mapping of Urban Areas Using EnMAP Data and Multioutput Support Vector Regression

Hyperspectral remote sensing data offer the opportunity to map urban characteristics in detail. Though, adequate algorithms need to cope with increasing data dimensionality, high redundancy between individual bands, and often spectrally complex urban landscapes. The study focuses on subpixel quantification of urban land cover compositions using simulated environmental mapping and analysis program (EnMAP) data acquired over the city of Berlin, utilizing both machine learning regression and classification algorithms, i.e., multioutput support vector regression (MSVR), standard support vector regression (SVR), import vector machine classifier (IVM), and support vector classifier (SVC). The experimental setup incorporates a spectral library and a reference land cover fraction map used for validation purposes. The library spectra were synthetically mixed to derive quantitative training data for the classes vegetation, impervious surface, soil, and water. MSVR and SVR models were trained directly using the synthetic mixtures. For IVM and SVC, a modified hyperparameter selection approach is conducted to improve the description of urban land cover fractions by means of probability outputs. Validation results demonstrate the high potential of the MSVR for subpixel mapping in the urban context. MSVR outperforms SVR in terms of both accuracy and computational time. IVM and SVC work similarly well, yet with lower accuracies of subpixel fraction estimates compared to both regression approaches.

Shapelet-based sparse image representation for landcover classification of hyperspectral data

This paper presents a novel sparse representation-based classifier for landcover mapping of hyperspectral image data. Each image patch is factorized into segmentation patterns, also called shapelets, and patch-specific spectral features. The combination of both is represented in a patch-specific spatial-spectral dictionary, which is used for a sparse coding procedure for the reconstruction and classification of image patches. Hereby, each image patch is sparsely represented by a linear combination of elements out of the dictionary. The set of shapelets is specifically learned for each image in an unsupervised way in order to capture the image structure. The spectral features are assumed to be the training data. The experiments show that the proposed approach shows superior results in comparison to sparse-representation based classifiers that use no or only limited spatial information and behaves competitive or better than state-of-the-art classifiers utilizing spatial information and kernelized sparse representation-based classifiers.

Tropical land use land cover mapping in Pará (Brazil) using discriminative Markov random fields and multi-temporal TerraSAR-X data

Abstract Remote sensing satellite data offer the unique possibility to map land use land cover transformations by providing spatially explicit information. However, detection of short-term processes and land use patterns of high spatial–temporal variability is a challenging task. We present a novel framework using multi-temporal TerraSAR-X data and machine learning techniques, namely discriminative Markov random fields with spatio-temporal priors, and import vector machines, in order to advance the mapping of land cover characterized by short-term changes. Our study region covers a current deforestation frontier in the Brazilian state Para with land cover dominated by primary forests, different types of pasture land and secondary vegetation, and land use dominated by short-term processes such as slash-and-burn activities. The data set comprises multi-temporal TerraSAR-X imagery acquired over the course of the 2014 dry season, as well as optical data (RapidEye, Landsat) for reference. Results show that land use land cover is reliably mapped, resulting in spatially adjusted overall accuracies of up to 79% in a five class setting, yet limitations for the differentiation of different pasture types remain. The proposed method is applicable on multi-temporal data sets, and constitutes a feasible approach to map land use land cover in regions that are affected by high-frequent temporal changes.