Publications by M.J. Collins

Speckle reduction for the forest mapping analysis of multi-temporal Radarsat-1 images

Wed, 07 Mar 2012

As the number of satellite-borne synthetic aperture radar (SAR) systems increases, both the availability and the length of multi-temporal (MT) sequences of SAR images have also increased. Previous research on MT SAR sequences suggests that they increase the classification accuracy for all applications over single date images. Yet the presence of speckle noise remains a problem and all images in the sequence must be speckle filtered before acceptable classification accuracy can be attained. Several speckle filters designed specifically for MT sequences have been reported in the literature. Filtering in the spatial domain, as is usually done, reduces the effective spatial resolution of the filtered image. MT speckle filters operate in both the spatial and temporal dimensions, thus the reduction in resolution is not likely to be as severe (although a comparison between MT and spatial filters has not been reported). While this advantage may be useful when extracting spatial features from the image sequence, it is not quite as apparent for classification applications. This research explores the relative performance of spatial and MT speckle filtering for a particular classification application: mapping boreal forest types. We report filter performance using the radiometric resolution as measured by the equivalent number of looks (NL), and classification performance as measured by the classification accuracy. We chose representative spatial and MT filters and found that spatial speckle filters offer the advantage of higher radiometric resolution and higher classification accuracy with lower algorithm complexity. Thus, we confirm that MT filtering offers no advantage for classification applications; spatial speckle filters yield higher overall performance.

Nonparametric feature selection and support vector machine for polarimetric SAR data classification

Wed, 07 Mar 2012

There are a wide range of SAR parameters that may be extracted from polarimetric SAR data. In very complex scenes like forests it is very useful to exploit the discriminative power offered by these features. Most of these features are of complex and sometimes unknown statistical properties. For this, the conventional feature selection algorithms cannot be applied. To account for this, a nonparametric separability measure was used as the evaluation function in the feature selection process. The measure is defined as the determinant of the between-class scatter matrix to the determinant of the sum of within-class scatter matrices. To improve the classification accuracy, the method was also used in the context of a class-based feature selection. The approach, first selects a feature subset for each class, train a SVM classifier on each selected feature subset and finally combines the outputs of the classifiers in a combination scheme. The Wishart classification algorithm is used as the reference method. Experimental results using Radarsat-2 data indicate that using the developed method improves the classification accuracy.

On the Use of Feature Selection for Classifying Multitemporal Radarsat-1 Images for Forest Mapping

Wed, 07 Mar 2012

As the number of satelliteborne SAR systems increases, both the availability and the length of multitemporal (MT) sequences of SAR images have also increased. Reported research with MT SAR sequences suggests that they increase the classification accuracy for all applications over single-date images. The length of the MT SAR sequences reported in the literature is still quite modest: on the order of six images. As the length of a sequence increases, the selection of images to use in a classification becomes important. The current practice is to add scenes chronologically, and some researchers have suggested that image selection does not affect classification accuracy. Our research explored the problem of image selection in MT SAR classification. We compared the chronological selection scheme with two feature selection algorithms: a very simple algorithm and a more complex class-based algorithm. We found that, while the simple feature selection algorithm was more efficient than chronological selection, yielding peak accuracy with few features, it saturated at the same accuracy as chronological selection. The more complex algorithm was significantly more accurate than chronological selection, even with just two features. Our results suggest that the use of a feature selection algorithm produces more efficient and more accurate classification results than chronological selection.

Estimating boreal forest species type with airborne polarimetric synthetic aperture radar

Wed, 07 Mar 2012

We have applied a non-parametric classifier (k nearest neighbour) to two calibrated orthogonal passes of airborne polarimetric synthetic aperture radar (POLSAR) image data over boreal forest for the purpose of discriminating canopy tree species of predefined stands. We found that a single classifier based on a single feature space (i.e. one set of POLSAR variables for all species) was less accurate than a hierarchical two-stage classifier that used different POLSAR variables for each species. We designed a two-stage classifier that first grouped stands into broad classes: pine, spruce and deciduous, and then classified each sample within the broad classes into individual species. We found that the most effective feature spaces had two or three dimensions. The two-stage classifier attained overall accuracies of between 60% and 75%.

We provide a first use of an equivalency test applied to remote-sensing classification. We use Lloyd's test of equivalency to find equivalent classifiers and thus infer informative POLSAR variables. The POLSAR variables that were most informative varied between the two passes and between the various elements of the hierarchical classifier. For the initial three-class classifier the most informative POLSAR variables were the two circular polarization ratios, several of Touzi's Stokes vector variables, HHVV coherence, several texture measures such as the variance of several scattering coefficients and the order parameter of the K-distribution and characteristics of the polarization signature pedestal. These results demonstrate that C-band POLSAR has great potential for mapping boreal forest cover either on its own or in concert with other geospatial data.

Analysing multitemporal SAR images for forest mapping

Wed, 07 Mar 2012

The objective of this paper is twofold: first, to presents a generic approach for the analysis of Radarsat-1 multitemporal data and, second, to presents a multi classifier schema for the classification of multitemporal images. The general approach consists of preprocessing step and classification. In the preprocessing stage, the images are calibrated and registered and then temporally filtered. The resulted multitemporally filtered images are subsequently used as the input images in the classification step. The first step in a classifier design is to pick up the most informative features from a series of multitemporal SAR images. Most of the feature selection algorithms seek only one set of features that distinguish among all the classes simultaneously and hence a limited amount of classification accuracy. In this paper, a class-based feature selection (CBFS) was proposed. In this schema, instead of using feature selection for the whole classes, the features are selected for each class separately. The selection is based on the calculation of JM distance of each class from the rest of classes. Afterwards, a maximum likelihood classifier is trained on each of the selected feature subsets. Finally, the outputs of the classifiers are combined through a combination mechanism. Experiments are performed on a set of 34 Radarsat-1 images acquired from August 1996 to February 2007. A set of 9 classes in a forest area are used in this study. Classification results confirm the effectiveness of the proposed approach compared with the case of single feature selection. Moreover, the proposed process is generic and hence is applicable in different mapping purposes for which a multitemporal set of SAR images are available.

The Synthetic Crown Recognition Model for Automatic Image Interpretation on UltraCamD Imagery

Mon, 01 Nov 2010