Anne H. Schistad Solberg

A Markov random field model for classification of multisource satellite imagery

A general model for multisource classification of remotely sensed data based on Markov random fields (MRF) is proposed. A specific model for fusion of optical images, synthetic aperture radar (SAR) images, and GIS (geographic information systems) ground cover data is presented in detail and tested. The MRF model exploits spatial class dependencies (spatial context) between neighboring pixels in an image, and temporal class dependencies between different images of the same scene. By including the temporal aspect of the data, the proposed model is suitable for detection of class changes between the acquisition dates of different images. The performance of the proposed model is investigated by fusing Landsat TM images, multitemporal ERS-1 SAR images, and GIS ground-cover maps for land-use classification, and on agricultural crop classification based on Landsat TM images, multipolarization SAR images, and GIS crop field border maps. The performance of the MRF model is compared to a simpler reference fusion model. On an average, the MRF model results in slightly higher (2%) classification accuracy when the same data is used as input to the two models. When GIS field border data is included in the MRF model, the classification accuracy of the MRF model improves by 8%. For change detection in agricultural areas, 75% of the actual class changes are detected by the MRF model, compared to 62% for the reference model. Based on the well-founded theoretical basis of Markov random field models for classification tasks and the encouraging experimental results in our small-scale study, the authors conclude that the proposed MRF model is useful for classification of multisource satellite imagery.

Multisource classification of remotely sensed data: fusion of Landsat TM and SAR images

Proposes a new method for statistical classification of multisource data. The method is suited for land-use classification based on the fusion of remotely sensed images of the same scene captured at different dates from multiple sources. It incorporates a priori information about the likelihood of changes between the acquisition of the different images to be fused. A framework for the fusion of remotely sensed data based on a Bayesian formulation is presented. First, a simple fusion model is given, and then the basic model is extended to take into account the temporal attribute if the different data sources are acquired at different dates. The performance of the model is evaluated by fusing Landsat TM images and ERS-1-SAR images for land-use classification. The fusion model gives significant improvements in the classification error rates compared to the conventional single-source classifiers. >

Oil Spill Detection in Radarsat and Envisat SAR Images

We present algorithms for automatic detection of oil spills in synthetic aperture radar (SAR) images. The algorithms consist of three main parts, namely: 1) detection of dark spots; 2) feature extraction from the dark spot candidates; and 3) classification of dark spots as oil spills or look-alikes. The algorithms have been trained on a large number of Radarsat and Envisat Advanced Synthetic Aperture Radar (ASAR) images. The performance of the algorithm is compared to manual and semiautomatic approaches in a benchmark study using 59 Radarsat and Envisat images. The algorithms can be considered to be a good alternative to manual inspection when large ocean areas are to be inspected

Remote Sensing of Ocean Oil-Spill Pollution

Oil spills on the sea surface are observed relatively often. Pollution due to either accidents or deliberate oily discharges from ships represents a serious threat to the marine environment. Operational oil spill monitoring is currently done using a combination of satellite monitoring and aircraft surveillance. The combined use of satellite-based synthetic aperture radar (SAR) images and aircraft surveillance flights is a cost-effective way to monitor oil spills in large ocean areas and catch the polluters. SAR images enable covering large areas, but aircraft observations are needed to prosecute the polluter, and in certain cases to verity the oil spill. Traditionally, oil spill detection is based on single polarization SAR images. Oil spills can be discriminated from look-alikes based on a set of features describing the contrast, shape, homogeneity, source, and surroundings of the slick. Good performance is reported for single-polarization oil spill detection, but in certain cases the oil slicks cannot be discriminated from biogenic films. In the recent years, a number of studies have shown that polarimetric SAR can improve the discrimination between oil slicks and biogenic films. Several features computed from dual-pol or quad-pol images have been proposed. These include both quad-pol features like polarimetric entropy and anisotropy, mean scattering angle, polarimetric span, conformity coefficient, as well as the dual-pol features standard deviation of the copolarized phase difference and the copolarized correlation coefficient. As dual-pol SAR imagery is now available on a regular basis from Cosmo Skymed and TerraSAR-X, and quad-pol data are available from RADARSAT-2, polarimetric SAR can now be utilized on a more regular basis. Optical data from sensors like Aqua MODIS and ENVISAT MERIS can be a useful supplement under certain cloud-free conditions.

Contextual data fusion applied to forest map revision

The use of a Markov random field model for multisource classification for map revision applications is investigated. A statistical model is presented, in which data from several remote sensing sensors is merged with spatial contextual information and a previous labeling of the scene from an existing thematic map to reach a consensus classification. The method is tested on two data sets for forest classification, and the classification performance is studied in terms of the effect of using remote sensing data from different sensors, the effect of spatial context, and the effect of using map data from previous surveys in the classification. It is shown that the use of a contextual classifier or an existing map of the area can have larger influence on the classification accuracy than using data from an additional sensor.

A bayesian approach to classification of multiresolution remote sensing data

Several earth observation satellites acquire image bands with different spatial resolutions, e.g., a panchromatic band with high resolution and spectral bands with lower resolution. Likewise, we often face the problem of different resolutions when performing joint analysis of images acquired by different satellites. This work presents models and methods for classification of multiresolution images. The approach is based on the concept of a reference resolution, corresponding to the highest resolution in the dataset. Prior knowledge about the spatial characteristics of the classes is specified through a Markov random field model at the reference resolution. Data at coarser scales are modeled as mixed pixels by relating the observations to the classes at the reference resolution. A Bayesian framework for classification based on this multiscale model is proposed. The classification is realized by an iterative conditional modes (ICM) algorithm. The parameter estimation can be based both on a training set and on pixels with unknown class. A computationally efficient scheme based on a combination of the ICM and the expectation-maximization algorithm is proposed. Results obtained on simulated and real satellite images are presented.

Oil Spill Detection in Hybrid-Polarimetric SAR Images

Oil spill detection in SAR images operating in a hybrid-polarimetric mode is examined. We propose and review several strategies for oil spill detection in hybrid-polarimetric SAR data. The retrieved measures are successfully applied to SAR data covering oil spill experiments outside Norway and the Deepwater Horizon incident in the Gulf of Mexico. It is shown that, under the assumption of a two-scale Bragg scattering model, a coherence measure may be recovered equally well from hybrid-polarimetric data, as for full-polarimetric data, and that this measure may be retrieved directly from the measurements without the need for any additional assumptions. The results show that low-wind lookalikes may be suppressed at the same time as the contrast of the oil spills is maintained using hybrid-polarimetric data and that multifeature images may be constructed to further enhance the oil spill detection performance. Due to the potential of wide swath widths, we conclude that hybrid-polarity is an attractive mode for future SAR-based oil spill monitoring.

Classifiers and Confidence Estimation for Oil Spill Detection in ENVISAT ASAR Images

An improved classification approach is proposed for automatic oil spill detection in synthetic aperture radar images. The performance of statistical classifiers and support vector machines is compared. Regularized statistical classifiers prove to perform the best on this problem. To allow the user to tune the system with respect to the tradeoff between the number of true positive alarms and the number of false positives, an automatic confidence estimator has been developed. Combining the regularized classifier with confidence estimation leads to acceptable performance.

Fast Hyperspectral Feature Reduction Using Piecewise Constant Function Approximations

The high number of spectral bands that are obtained from hyperspectral sensors, combined with the often limited ground truth, solicits some kind of feature reduction when attempting supervised classification. This letter demonstrates that an optimal constant function representation of hyperspectral signature curves in the mean square sense is capable of representing the data sufficiently to outperform, or match, other feature reduction methods such as principal components transform, sequential forward selection, and decision boundary feature extraction for classification purposes on all of the four hyperspectral data sets that we have tested. The simple averaging of spectral bands makes the resulting features directly interpretable in a physical sense. Using an efficient dynamic programming algorithm, the proposed method can be considered fast.

Structured Gaussian Components for Hyperspectral Image Classification

The large number of bands in hyperspectral images leads to a large number of parameters to estimate. It has been argued in the literature that class-conditional distributions of hyperspectral images are non-Gaussian; thus, multiple components might be needed to describe the classes accurately. In this paper, we propose to represent the Gaussian components in the classifier with a smaller number of parameters by allowing some or all component distributions to share eigenstructure by decomposing the covariance matrix Sigmak of each of the k components into a product of three parameters, namely: 1) scalar lambdak measuring volume; 2) diagonal matrix Ak of normalized eigenvalues measuring shape; and 3) matrix of eigenvectors Dk measuring orientation. Any combination of these parameters can be common for any subset of the covariance matrices, allowing a flexible set of possible configurations that can be used to approximate the true covariance using fewer parameters. A simple bottom-up algorithm for searching for possible parameter-sharing models is developed. Experiments on three data sets were performed: one concerned with woodland classification and two on urban mapping. Results from these experiments indicate that the method outperforms conventional classifiers and performs comparably with state-of-the-art classifiers such as support vector machines