Morton J. Canty

Visualization and unsupervised classification of changes in multispectral satellite imagery

The statistical techniques of multivariate alteration detection, minimum/maximum autocorrelation factors transformation, expectation maximization and probabilistic label relaxation are combined in a unified scheme to visualize and to classify changes in multispectral satellite data. The methods are demonstrated with an example involving bitemporal LANDSAT TM imagery.

Playing for time: A sequential inspection game

Abstract Inspections for timely detection of illegal activity on a finite, closed time interval and subject to first and second kind errors are modelled as a sequential, two-person game. The utilities of the players, inspector and inspectee, are assumed to be linear in the detection time with time-independent false alarm costs. Sets of Nash equilibria are obtained in which the inspectee behaves illegally or legally with probability one.

Inspection games in arms control

An inspection game is a mathematical model of a situation in which an inspector verifies the adherence of an inspectee to some legal obligation, such as an arms control treaty, where the inspectee may have an interest in violating that obligation. The mathematical analysis seeks to determine an optimal inspection scheme, ideally one which will induce legal behavior, under the assumption that the potential illegal action is carried out strategically; thus a non-cooperative game with two players, inspector and inspectee, is defined. Three phases of development in the application of such models to arms control and disarmament may be identified. In the first of these, roughly from 1961 through 1968, studies that focused on inspecting a nuclear test ban treaty emphasized game theory, with less consideration given to statistical aspects associated with data acquisition and measurement uncertainty. The second phase, from 1968 to about 1985, involves work stimulated by the Treaty on the Non-Proliferation of Nuclear Weapons (NPT). Here, the verification principle of material accountancy came to the fore, along with the need to include the formalism of statistical decision theory within the inspection models. The third phase, 1985 to the present, has been dominated by challenges posed by such far-reaching verification agreements as the Intermediate Range Nuclear Forces Agreement (INF), the Treaty on Conventional Forces in Europe (CFE) and the Chemical Weapons Convention (CWC), as well as perceived failures of the NPT system in Iraq and North Korea. In this connection, the interface between the political and technical aspects of verification is being examined from the game-theoretic viewpoint.

Improving Change Detection Results of IR-MAD by Eliminating Strong Changes

This letter examines the effect of the prior elimination of strong changes on the results of change detection in bitemporal multispectral images using the previously published iteratively reweighted multivariate alteration detection (IR-MAD) method. An initial change mask is calculated by identifying strong changes between two images. By using the mask and hence eliminating the strong changes from the analysis, the IR-MAD method is able to identify a better no-change background. This effect is demonstrated on a multitemporal Landsat Enhanced Thematic Mapper Plus data set from an agricultural region in Germany with substantial improvement in the results even for the scenes which have a large number of changes.

Boosting a fast neural network for supervised land cover classification

It is demonstrated that the use of an ensemble of neural networks for routine land cover classification of multispectral satellite data can lead to a significant improvement in classification accuracy. Specifically, the AdaBoost.M1 algorithm is applied to a sequence of three-layer, feed-forward neural networks. In order to overcome the drawback of long training time for each network in the ensemble, the networks are trained with an efficient Kalman filter algorithm. On the basis of statistical hypothesis tests, classification performance on multispectral imagery is compared with that of maximum likelihood and support vector machine classifiers. Good generalization accuracies are obtained with computation times of the order of 1h or less. The algorithms involved are described in detail and a software implementation in the ENVI/IDL image analysis environment is provided.

Kernel principal component analysis for change detection

Principal component analysis (PCA) is often used to detect change over time in remotely sensed images. A commonly used technique consists of finding the projections along the two eigenvectors for data consisting of two variables which represent the same spectral band covering the same geographical region acquired at two different time points. If change over time does not dominate the scene, the projection of the original two bands onto the second eigenvector will show change over time. In this paper a kernel version of PCA is used to carry out the analysis. Unlike ordinary PCA, kernel PCA with a Gaussian kernel successfully finds the change observations in a case where nonlinearities are introduced artificially.

Automation of change detection procedures for nuclear safeguards-related monitoring purposes

Against the background of nuclear safeguards applications using commercially available satellite imagery, a two-steps attempt for change detection and analysis was realized in general. Beginning with the wide-area monitoring on the basis of medium-resolution satellite data for the pre-scanning of significant changes within the nuclear-related locations, the areas of interest could then be explicitly analyzed by change detection and analysis methods using high-resolution satellite data. The change pixels were detected by using the multivariate alteration detection (MAD) transformation, producing a set of mutually orthogonal difference images (the so-called MAD variates). The decision thresholds for the change pixels were set by applying a probability mixture model to the MAD variates based on an EM algorithm. By means of eCognition a second, object-oriented procedure was implemented in order to create an automated workflow for the multiscale extraction of the (change) objects and (change) features for the subsequent post-classification of the areas of interest. Regarding the necessity of automation for extensive monitoring tasks the processing aspects of standardization and transferability took the centre stage of the investigations.

Multi- and hyperspectral remote sensing change detection with generalized difference images by the IR-MAD method

Change detection methods for multi- and hyper- variate data aim at identifying differences in data acquired over the same area at different points in time. In this con- tribution an iterative extension to the multivariate alteration detection (MAD) transformation for change detection is sketched and applied. The MAD transformation is based on canonical correlation analysis (CCA), which is an established technique in multivariate statistics. The extension in an iterative scheme seeks to establish an increasingly better background of no-change against which to detect change. This is done by putting higher weights on observations of no-change in the calculation of the statistics for the CCA. The differences found may be due to noise or differences in (atmospheric etc.) conditions at the two acquisition time points. To prevent a change detection method from detecting uninteresting change due to noise or arbitrary spurious differences the application of regularization, also known as penalization, and other types of robustification of the change detection method may be important especially when applied to hyperspectral data. Among other things results show that the new iterated scheme does give a better no-change background against which to detect change than the original, non-iterative MAD method and that the IR-MAD method depicts the change detected in less noisy components. I. INTRODUCTION This contribution focuses on construction of more gen- eral difference images than simple differences in multivariate change detection. This is done via an iterated version (1) of the canonical correlation analysis (CCA) (2) based multivariate alteration detection (MAD) method (3) that could, moreover, be combined with an expectation-maximization (EM) based method for determining thresholds for differentiating between change and no-change in the difference images, and for estimating the variance-covariance structure of the no-change observations (4), (5). The variances can be used to estab- lish a single change/no-change image based on the general multivariate difference image. The resulting imagery from MAD based change detection is invariant to linear and affine transformations of the input including, e.g., affine corrections to normalize data between the two acquisition time points. This is an enormous advantage over other multivariate change detection methods. The resulting single change/no-change image can be used to establish both change regions and to extract observations with which a fully automated orthogonal regression analysis based normalization of the multivariate data between the two points in time can be developed (6). Results (not shown here) from partly simulated multivariate data indicate an improved performance of the iterated scheme over the original MAD method (1). Also, a few comparisons with established methods for calculation of robust statistics for the CCA indicate that the scheme suggested here performs better, see also (7). Regularization issues typically important in connection with the analysis of hyperspectral data are dealt with in (8)-(10) and briefly mentioned here.

Unsupervised classification of changes in multispectral satellite imagery

The statistical techniques of multivariate alteration detection, maximum autocorrelation factor transformation, expectation maximization, fuzzy maximum likelihood estimation and probabilistic label relaxation are combined in a unified scheme to classify changes in multispectral satellite data. An example involving bitemporal LANDSAT TM imagery is given.

Kernel principal component and maximum autocorrelation factor analyses for change detection

Kernel versions of the principal components (PCA) and maximum autocorrelation factor (MAF) transformations are used to postprocess change images obtained with the iteratively re-weighted multivariate alteration detection (MAD) algorithm. It is found that substantial improvements in the ratio of signal (change) to background noise (no change) can be obtained especially with kernel MAF.