Your search

Constraint-based spatial reasoning problems frequently arise in the area of military mission planning. In this domain, mission planners employ complex criteria, which may include numeric and optimization constraints in addition to logical constraints and rules, to develop engineering construction and resource deployment plans. Automated planning aid systems for the military must have the capability to represent the various types of constraints used in human decision-making and must be able to provide efficient and optimal or near optimal solutions to the resulting constraint satisfaction problems. This paper describes a methodology for transforming constraint satisfaction problems into nonlinear optimization problems and for solving the resulting optimization problems using a hybrid neural network/genetic algorithm procedure. The method is applied to illustrative problems which employ different types of constraints for determination of future construction sites. The results of the experiments demonstrate the potential of this methodology for finding feasible and optimal solutions to nonlinear optimization problems. © 1994, Elsevier B.V.

The primary objective of this project is to define a methodology to depict the motion of deep convective cloud systems as observed form satellite imagery. These clouds are defined as clusters of pixels with Cloud Top Pressure (IPC) <EQ 440 millibars and Cloud Optical Thickness (TAU) >= 23 which are high in the atmosphere and sufficiently thick to produce significant rainfall. Clouds are one of the major factors in understanding the earth's climate. Evaluating cloud motion is important in understanding atmospheric dynamics and visualizations are vital because they provide a good way to observe change. IPC and TAU values have been collected for April of 1989 from the International Satellite Cloud Climatology Project, low resolution database for the northern latitudes between 30 and 60 degrees. Each of the 240 IPC and 240 TAU images consisted of 12 rows and 144 columns with each pixel representing a 280 km square on the globe collected in three-hour intervals. Individual images were color coded according to land, sea and clouds before being put into motion. Six animations have been produced which start with the original images, progress to include daily composite images and culminate with a collage. Animations of the original images have the advantage of relatively short intervals between still frames but have many undefined pixels, which are eliminated in the composites. The results of this project can serve as an example of how to improve the visualization of time varying image sequences.

Backpropagation neural networks are applied to the problem of characterization of ultrasonic image texture to detect abnormalities in tissue texture which are indicative of liver disease. Twenty-one texture features were extracted from regions of interest in digitized ultrasonic images. A feature subset, identified by a stepwise selection process, formed the sample input to the networks together with the physician-supplied diagnosis. The classification performance of the backpropagation network is evaluated using a jackknife testing procedure. The performance of the networks is compared with results obtained from linear discriminant analysis and logistic regression techniques. © Springer-Verlag Berlin Heidelberg 1995.

Hybrid knowledge bases (HKBs), proposed by Nerode and Subrahmanian, provide a uniform theoretical framework for dealing with the mixed data types and multiple reasoning modes required for solving logical deployment problems. Algorithms based on mixed integer linear programming techniques have been developed for the syntactic subset of HKBs corresponding to function-free Prolog-like logic programs. In this study, we examine the ability of neural networks to solve a more comprehensive set of problems expressed within the hybrid knowledge base framework. The objective of this research is to design and implement a nonlinear optimization procedure for solving extended logic programs with neural networks. We focus upon two types of extensions which are typically required in the formulation of logical deployment problems. The first type of extension, which we shall refer to as a Type I extension, consists of embedding numerical and geometric constraints into logic programs. The second type of extension, which we shall call a Type II extension, consists of incorporating optimization problems into logic clauses. © 1993 SPIE. All rights reserved.

Accurate identification and tracking of synoptic-scale storm systems in the northern midlatitudes is important for understanding the structure and movement of the midlatitude cloud field which plays a major role in climate change. In this paper, a hybrid neural network/genetic algorithm (NN/GA) approach is presented that analyzes the behavior of storm systems from one time frame to the next. The goal of the hybrid neural network algorithm is to improve classifier output by reducing the number of infeasible solutions using constraint optimization techniques. The input to the hybrid neural network algorithm is the output from a traditional backpropagation neural network. The hybrid NN/GA analyzes the backpropagation neural network output for logical consistencies and makes changes to the classification results based on strength of neural network classifications and satisfaction of logical constraints. The results are compared with classification results obtained using linear discriminant analysis, k-nearest neighbor rule, and backpropagation neural network techniques.

The purpose of this study was to compare the classification capabilities of the backpropagation algorithm and linear discriminant analysis for detecting liver metastisis using image texture features obtained from ultrasonic images of the liver. Twenty-one quantitative parameters were obtained from 134 regions of interest of equal size. The images were collected by the same radiologist on the same imager with the controls adjusted for variations in patient body size so as to produce images of consistent quality. Quantitative features were divided so that 13 were first-order statistics, 6 were second-order statistics, and 2 were image gradient parameters. The same features were processed by both the backpropagation algorithm and linear discriminant analysis using `jack-knife'' testing and the results of each computer- generated classification was compared to the supplied diagnosis in an effort to determine which method could best identify patterns. For this particular application, the backpropagation neural network was found to have slightly superior classification results (87) than linear discriminant analysis (83).

The objective of this study is to compare statistical and unsupervised neural network techniques for determination of correspondences between storm system regions extracted from sequences of satellite images. Analysis was applied to the International Satellite Cloud Climatology Project (ISCCP) low resolution D1 database for selected storm systems during the period April 5 - 9, 1989. Cloud top pressure was used to delineate regions of interest and cloud optical thickness combined with spatial location was used to track regions throughout a given time sequence. The ability of the k-nearest neighbor classifier and of self-organizing maps to determine correspondences between storm regions was assessed. The two techniques generally yielded similar associations between regions of interest throughout the time sequence. Differences in final tracking results between the two techniques occurred primarily as a result of differences in the collections of points from a region in a time step t<SUB>2</SUB> that corresponded to a region in an earlier time step t<SUB>1</SUB>. The tracking results were also compared to the results obtained at the NASA Goddard Institute for Space Studies using sea level pressure data from the National Meteorological Center (NMC). For the storm systems investigated in this study, the storm tracks exhibited the same general tracking behavior with expected variations between cloud system storm centers and low sea level pressure centers.

The primary objective of this research is the development and testing of neural network models for two fundamental computer vision tasks: edge/line detection and texture analysis. In order to test the ability of the neural network models to detect patterns in images we used both remote sensing data and medical imagery. Neural network models for edge and line detection were used to detect geological lineaments in Landsat data. Neural network models for the analysis of image texture variations were used on ultrasonic images to distinguish patients with normal liver scans from patients with diffuse liver disease. 1.

An evolutionary system was developed for generation of complete tracks of northern midlatitude synoptic-scale storm systems based on optical flow and cloud motion analyses of global satellite-based datasets produced by the International Satellite Cloud Climatology Project (ISCCP). The tracking results were compared with low sea level pressure anomaly (SLPA) tracks obtained from the NASA Goddard Institute for Space Studies (GISS). The SLPA tracks were produced at GISS by analysis of meteorological, ground-based National Center for Environmental Prediction (NCEP) datasets. Results from the evolutionary system were also compared with results from using (a) the k-nearest neighbor rule (k-NN) and (b) self-organizing maps (SOM) to determine correspondences between consecutive locations within a track. The consistency of our evolutionary storm tracking results with the behavior of the low sea level pressure anomaly tracks, the ability of our evolutionary system to generate and evaluate complete tracks, and the close comparison between the results obtained by the evolutionary, k-NN, and SOM analyses of the ISCCP-derived datasets at tracking steps in which proximity or optical flow information sufficed to determine movement, demonstrate the applicability and the potential of evolutionary systems for tracking midlatitude storm systems through low-resolution ISCCP cloud product datasets.

Cloud analyses provide information which is vital to the detection, understanding and prediction of meteorological trends and environmental changes. This paper compares statistical, neural network and genetic algorithm methods for recognition and tracking of midlatitude storm clouds in sequences of low-resolution cloud-top pressure data sets. Regions of interest are identified and tracked from one image frame to the next consecutive frame in an eight-frame sequence. Classification techniques are used to determine the relationships between regions of interest in consecutive time frames. A genetic algorithm procedure is then used to revise classifier outputs to ensure that consistency constraints are not violated. © 1997 Elsevier Science B.V.

The objective of this research is to automate the classification of the temporal behavior of storm cloud systems based on measurements derived from consecutive satellite images. The motivation behind this study is to develop improved descriptions of cloud dynamics which can be used in general circulation models for prediction of global climate change. Analysis was applied to the International Satellite Cloud Climatology Project (ISCCP) low resolution cloud top pressure database for the first six days in April, 1989. A total of 296 midlatitude storm cloud components were tracked between consecutive 3-hour time frames. For each pair of components, temporal correspondence events were classified as either 1.) direct, 2.) merge, 3.) split, or 4.) reject. The reject class, which was used primarily to categorize pairs of unrelated systems, included storm cloud system dissipation and creation as well. Statistical, neural network, and evolutionary techniques were developed for finding solutions to the storm cloud correspondence problem. Evolutionary techniques applied to the problem consisted of 1.) a constraint-handling hybrid evolutionary technique and 2.) a genetic local search algorithm. The results demonstrate the potential of evolutionary techniques to yield meteorologically-feasible solutions, given appropriate constraints, to the two-frame storm tracking problem. © 1998 SPIE. All rights reserved.

In the past fractal dimension has often been computed using a stochastic approach based on a random walk process, which has been found to be very time consuming. More recently, mathematical morphology has been used to compute the fractal dimension in a more timely fashion. This paper describes how the fractal dimension computed using mathematical morphology can be used in the texture analysis of ultrasonic imagery. The discriminatory ability of the fractal dimension as a pattern recognition feature is evaluated and compared to more traditional parameters. This analysis includes comparisons with statistical features in which each parameter is treated as an independent variable and in which interactions between those variables are evaluated. Pattern recognition techniques include Stepwise Discriminant Analysis, Linear Discriminant Analysis, and Nearest Neighbor Analyisis in addition to Backpropagation Neural Network Classifiers. Our results identify the fractal dimension as one of the most important parameters for distinguishing between normal and abnormal livers. In this study, consisting of 186 images, a significant statistical difference was found for both the mean and standard deviation of the fractal dimension between the normal and abnormal groups using parametric and nonparametric statistical techniques. © 1993 SPIE. All rights reserved.

Temporal analysis has been applied to a sequence of cloud top pressure (CTP) images and cloud optical thickness (TAU) images stored in the International Satellite Cloud Climatology Project (ISCCP) D1 database located at the NASA Goddard Institute for Space Studies (GISS). Each pixel in the D1 data set has a resolution of 2.5 degrees or 280 kilometers. These images were collected in consecutive three-hour intervals for the entire month of April 1989. The primary objective of this project was to develop a sequence of storm tracks from the satellite images to follow the formation, progression and dissipation of storm systems over time. Composite images where created by projecting ahead in time and substituting the first available valid pixel for missing data and a variety of CTP and TAU cut-off values were used to identify regions of interest. Region correspondences were determined from one time frame to another yielding the coordinates of storm centers. These tracks were compared to storm tracks computed from sea level pressure data obtain from the National Meteorological Center (NMC) for the same time period. The location of sea level storm center provides insight as to whether storms have occurred anywhere in a region and can be helpful in determining the presence or absence of storms in a general geographic region.

Over the past several years we have been interested in the supervised classification of ultrasonic images of the liver based on quantitative texture features. Our most recent efforts are concerned with the inclusion of features computed from Markov random fields. After adding four such features to our existing model containing 17 features, we employed stepwise discriminant analysis to identify the features that could best discriminate among 184 previously classified normal and abnormal ultrasonic images. Three of the four features derived from Markov random field models were identified by stepwise discriminant analysis as being good discrimination along with 6 existing features. From these results we constructed a backpropagation neural network with an input layer consisting of 9 nodes. We found that this new model yielded slightly better results when compared to earlier models. Our most recent results yielded a sensitivity of 81%, a specificity of 77% and an overall accuracy of 79%.

One of the major problems in the development of computer- A ssisted systems for geologic mapping is how to individualize the system to meet user needs. Ideally, the system should be responsive to specifications of desired types of output structures. Also the system should be able to incorporate the user's knowledge of regional characteristics into the feature extraction/selection and classification components. Automatic techniques for classification of remote sensing data typically require relatively large, labeled training sets which are well-organized with respect to the desired mapping between input and output patterns. The present paper focuses on the feature extraction/selection component of the system. Kohonen self-organizing feature maps in conjunction with image processing procedures for linear feature extraction are used for explorative data analysis, feature selection, and construction of exemplar patterns. The results of training Kohonen feature maps with different pattern sets and different feature combinations provide insight into the nature of pattern relationships which enables the user to develop sets of positive and negative training patterns for the classification component. © 1992 SPIE. All rights reserved.

Backpropagation neural networks have been developed for detection of geological lineaments in the Landsat Thematic Mapper (TM) imagery of the Canadian Shield using edge images as input and digitized lineament maps as the desired output. Lineament detection is a challenging problem for traditional image processing and pattern recognition techniques. Many linear features observable in geological image data do not represent lineaments, and the presence and extent of lineaments must be inferred from contextual information. In order to compare the ability of neural networks and conventional classifiers to recognize lineaments prior to performing edge/line element grouping operations, various gradient and curvature features are extracted from the image data set. Selected features from this group formed the inputs to backpropagation neural networks, linear discriminant classifiers, and nearest-neighbor classifiers. The neural network results were compared with the results obtained using conventional classifiers for sample training and test sets. The trained neural network was then applied to the edge image to mask out those edge points which had been classified as non- lineament points.