Dirk Borghys

Fusion of PolSAR and PolInSAR data for land cover classification.

Abstract The main research goal of this study is to investigate the complementarity and fusion of different frequencies (L- and P-band), polarimetric SAR (PolSAR) and polarimetric interferometric (PolInSAR) data for land cover classification. A large feature set was derived from each of these four modalities and a two-level fusion method was developed: Logistic regression (LR) as ‘feature-level fusion’ and the neural-network (NN) method for higher level fusion. For comparison, a support vector machine (SVM) was also applied. NN and SVM were applied on various combinations of the feature sets. The results show that for both NN and SVM, the overall accuracy for each of the fused sets is better than the accuracy for the separate feature sets. Moreover, that fused features from different SAR frequencies are complementary and adequate for land cover classification and that PolInSAR is complementary to PolSAR information and that both are essential for producing accurate land cover classification.

Supervised feature-based classification of multi-channel SAR images

This paper describes a new method for a feature-based supervised classification of multi-channel SAR data. Classic feature selection and classification methods are inadequate due to the diverse statistical distributions of the input features. A method based on logistic regression (LR) and multinomial logistic regression (MNLR) for separating different classes is therefore proposed. Both methods, LR and MNLR, are less dependent on the statistical distribution of the input data. A new spatial regularization method is also introduced to increase consistency of the classification result. The classification method was applied to a project on humanitarian demining in which the relevant classes were defined by experts of a mine action center. A ground survey mission collected learning and validation samples for each class. Results of the proposed classification methods are shown and compared to a maximum likelihood classifier.

Comparative evaluation of hyperspectral anomaly detectors in different types of background

Anomaly detection (AD) in hyperspectral data has received a lot of attention for various applications. The aim of anomaly detection is to detect pixels in the hyperspectral data cube whose spectra differ significantly from the background spectra. Many anomaly detectors have been proposed in literature. They differ by the way the background is characterized and by the method used for determining the difference between the current pixel and the background. The most well-known anomaly detector is the RX detector that calculates the Mahalanobis distance between the pixel under test (PUT) and the background. Global RX characterizes the background of the complete scene by a single multi-variate normal distribution. In many cases this model is not appropriate for describing the background. For that reason a variety of other anomaly detection methods have been developed. This paper examines three classes of anomaly detectors: sub-space methods, local methods and segmentation-based methods. Representative examples of each class are chosen and applied on a set of hyperspectral data with different backgrounds. The results are evaluated and compared.

Edge and line detection in polarimetric SAR images

A scheme for detecting edges and lines in multichannel SAR images is proposed. The line detector is constructed from the edge detector. The latter is based on multivariate statistical hypothesis tests applied to log-intensity SAR images. The raw results are vectorized by a traditional bright line extraction process. The scheme is illustrated by extracting dark linear structures on various full-polarimetric SAR images.

Multilevel data fusion for the detection of targets using multispectral image sequences

An approach is presented to the long range automatic detec- tion of vehicles, using multisensor image sequences. The method is tested on a database of multispectral image sequences, acquired under diverse operational conditions. The approach consists of two parts. The first part uses a semisupervised approach, based on texture parameters, for detecting stationary targets. For each type of sensor one learning image is chosen. Texture parameters are calculated at each pixel of the learning images and are combined using logistic regression into a value that represents the conditional probability that the pixel belongs to a target given the texture parameters. The actual detection algorithm ap- plies the same combination to the texture features calculated on the remainder of the database (test images). When the results of this feature-level fusion are stored as an image, the local maxima correspond to likely target positions. These feature-level-fused images are calcu- lated for each sensor. In a sensor fusion step, the results obtained per sensor are then combined again. Region growing around the local maxima is then used to detect the targets. The second part of the algo- rithm searches for moving targets. To detect moving vehicles, any mo- tion of the sensor must be detected first. If sensor motion is detected, it is estimated using a Markov random field model. Available prior knowl- edge about the sensor motion is used to simplify the motion estimation. The estimate is used to warp past images onto the current image in a temporal fusion approach and moving targets are detected by threshold- ing the difference between the original and warped images. Decision level fusion combines the results from both parts of the algorithm.

Hyperspectral anomaly detection: comparative evaluation in scenes with diverse complexity

Anomaly detection (AD) in hyperspectral data has received a lot of attention for various applications. The aim of anomaly detection is to detect pixels in the hyperspectral data cube whose spectra differ significantly from the background spectra. Many anomaly detectors have been proposed in the literature. They differ in the way the background is characterized and in the method used for determining the difference between the current pixel and the background. The most well-known anomaly detector is the RX detector that calculates the Mahalanobis distance between the pixel under test (PUT) and the background. Global RX characterizes the background of the complete scene by a singlemultivariate normal probability density function. Inmany cases, this model is not appropriate for describing the background. For that reason a variety of other anomaly detection methods have been developed. This paper examines three classes of anomaly detectors: subspace methods, local methods, and segmentation-based methods. Representative examples of each class are chosen and applied on a set of hyperspectral data with diverse complexity. The results are evaluated and compared.

A multivariate contour detector for high-resolution polarimetric SAR images

This article presents a new method for contour detection in high-resolution polarimetric SAR images. The method is based on multivariate statistics of speckle in homogeneous regions in a SAR image and uses a hypothesis test for the difference in variance between two adjacent regions to find the contours. The detector is directly applied to the single-look complex polarimetric SAR image. A preprocessor which is also based on multivariate statistics, is used to focus the attention of the detector on potentially non-homogeneous regions within the image. Results of applying the contour detector on L-band polarimetric SAR images are also presented.

Hyperspectral anomaly detection: A comparative evaluation of methods

Anomaly detection in hyperspectral data has received a lot of attention for various applications. The aim of anomaly detection is to detect pixels in the hyperspectral datacube whose spectra differ significantly from the background spectra. In anomaly detection no prior knowledge about the target is assumed. Anomaly detection methods in general estimate the spectra of the background (locally or globally) and then detect anomalies as pixels with a large spectral distance w.r.t. the determined background spectra. Many types of anomaly detectors have been proposed in literature, each depending on several parameters. The aim of this paper is to compare the results of different types of anomaly detection when they are applied to scenes with different complexity: urban scenes with different complexity and rural scenes with sub-pixel anomalies. This paper only considers hyperspectral data in the VNIR and SWIR part of the EM spectrum (λ = 0.4–2.5μm).

Automatic detection of built-up areas in high-resolution polarimetric SAR images

Abstract In this paper a method for automatic detection of built-up areas in high-resolution polarimetric SAR images is presented. A feature-based approach was used. Most of the features are based on statistical properties of built-up areas in SAR images. One feature is based on the isotropic spatial distribution of small uniform regions within built-up areas. Introducing this feature allows one to avoid false alarms due to the presence of edges. The features are fused using logistic regression. Results of applying the method on an L-Band fully polarised SAR image with a spatial resolution of 1.5 m are shown.

Contour detection in high-resolution polarimetric SAR images

Automatic contour detection in SAR images is a difficult problem due to the presence of speckle. Several detectors exploiting the statistics of speckle in uniform regions have been already presented in literature. However, these were mainly applied to multi-look low-resolution imagery. This paper describes two new CFAR contour detectors for high-resolution single-look polarimetric SAR images. They are based on multi-variate statistical hypothesis tests. Failing of the test indicates the presence of an edge. A test for difference in means on log-intensity images and difference in variance on complex (SLC) images are used. Both tests take into account the interchannel covariance matrix which makes them a powerful tool for contour detection in multi-channel SAR images. Spatial correlation jeopardizes the CFAR character of the detectors. This problem is often neglected. In this paper its influence on the detectors is studied and eliminated. The localisation of detected edges is improved using a directional morphological filter. Different methods to fuse the results of the two detectors are explored and compared. Results obtained on a polarimetric L-band E-SAR image are presented. Most contours are well detected. Narrow lines on a uniform background remain undetected. Although the detector was developed to detect edges only between uniform areas, it also detects edges between textured and uniform areas.