Dale E. Nelson

Entropy-based optimum test points selection for analog fault dictionary techniques

An efficient method to select an optimum set of test points for dictionary techniques in analog fault diagnosis is proposed. This is done by searching for the minimum of the entropy index based on the available test points. First, the two-dimensional integer-coded dictionary is constructed whose entries are measurements associated with faults and test points. The problem of optimum test points selection is, thus, transformed to the selection of the columns that isolate the rows of the dictionary. Then, the likelihood for a column to be chosen based on the size of its ambiguity set is evaluated using the minimum entropy index of test points. Finally, the test point with the minimum entropy index is selected to construct the optimum set of test points. The proposed entropy-based method to select a local minimum set of test points is polynomial bounded in computational cost. The comparison between the proposed method and other reported test points selection methods is carried out by statistical experiments. The results indicate that the proposed method more efficiently and more accurately finds the locally optimum set of test points and is practical for large scale analog systems.

A Mathematical Foundation for Improved Reduct Generation in Information Systems

Abstract. When data sets are analyzed, statistical pattern recognition is often used to find the information hidden in the data. Another approach to information discovery is data mining. Data mining is concerned with finding previously undiscovered relationships in data sets. Rough set theory provides a theoretical basis from which to find these undiscovered relationships. We define a new theoretical concept, strong compressibility, and present the mathematical foundation for an efficient algorithm, the Expansion Algorithm, for generation of all reducts of an information system. The process of finding reducts has been proven to be NP-hard. Using the elimination method, problems of size 13 could be solved in reasonable times. Using our Expansion Algorithm, the size of problems that can be solved has grown to 40. Further, by using the strong compressibility property in the Expansion Algorithm, additional savings of up to 50% can be achieved. This paper presents this algorithm and the simulation results obtained from randomly generated information systems.

Iterated wavelet transformation and signal discrimination for HRR radar target recognition

This paper explores the use of wavelets to improve the selection of discriminant features in the target recognition problem using high range resolution (HRR) radar signals in an air to air scenario. We show that there is statistically no difference among four different wavelet families in extracting discriminatory features. Since similar results can be obtained from any of the four wavelet families and wavelets within the families, the simplest wavelet (Haar) should be used. We use the box classifier to select the 128 most salient pseudo range bins and then apply the wavelet transform to this reduced set of bins. We show that by iteratively applying this approach, the classifier performance is improved. We call this the iterated wavelet transform . The number of times the feature reduction and transformation can be performed while producing improved classifier performance is small and the transformed features are shown to quickly cause the performance to approach an asymptote.

Advanced feature selection methodology for automatic target recognition

The paper investigates independent feature selection as used in neural networks for solving classification problems. Radial basis functions and wavelet transforms are used to preprocess the input data. A class of nonorthogonal classifiers is defined and their properties are investigated. It is demonstrated that nonorthogonal classifiers perform better than the orthogonal ones. Feature selection using mutual information is also investigated. Independence of features based on the information content is defined and used to select features for synthesis of ontogenic neural networks. Simulation results using synthetically generated radar returns showed promise for automatic target recognition.

High range resolution radar signal classification a partitioned rough set approach

In automatic target recognition (ATR) systems there are advantages to developing classifiers based on a portion of the signal. A partitioning technique is introduced in this paper that allows rough set theory to be applied to real-world size problems. Rough set theory (RST) is an emerging concept for determining features and then classifiers from a training data set. RST guarantees that once the data has been labeled all possible classifiers (based on that labeling) can be generated. There are multiple classifiers for each signal partition and multiple partitions for each signal. Classifiers based on a single reduct (classifier) or one partition do not perform well enough to be useful. We fuse all the reducts from all the partitions into one classifier. This fusion of partitioned reducts yields a synergistic result that produces a classifier with a high probability of declaration and good probability of correct classification.

Wavelet Transformation and Signal Discrimination for HRR Radar Target Recognition

This paper explores the use of wavelets to improve the selection of discriminant features in the target recognition problem using High Range Resolution (HRR) radar signals in an air to air scenario. We show that there is statistically no difference between four different wavelet families in extracting discriminatory features. Since similar results can be obtained from any of the four wavelet families and wavelets within the families, the simplest wavelet (Haar) should be used. We further show that a simple box classifier can be constructed from the extracted features and that any feature that classifies four or less training signals can be removed from the classifier without a statistically significant difference in the classifier performance. We use the box classifier to select the 128 most salient pseudo range bins and then apply the wavelet transform to this reduced set of bins. We show that by iteratively applying this approach, classifier performance is improved. The number of times the feature reduction and transformation can be performed while producing improved classifier performance is small and the transformed features are shown to quickly cause the performance to approach an asymptote.