Mary M. Moya

Network constraints and multi-objective optimization for one-class classification

Abstract This paper introduces a constrained second-order network with a multiple objective learning algorithm that forms closed hyperellipsoidal decision boundaries for one-class classification. The network architecture has uncoupled constraints that give independent control over each decision boundarys size, shape, position, and orientation. The architecture together with the learning algorithm guarantee the formation of positive definite eigenvalues for closed hyperellipsoidal decision boundaries. The learning algorithm incorporates two criteria, one that seeks to minimize classification mapping error and another that seeks to minimize the size of the decision boundaries. We consider both additive combinations and multiplicative combinations of the individual criteria, and we present empirical evidence for selecting functional forms of the individual objectives that are bounded and normalized. The resulting multiple objective criterion allows the decision boundaries to increase or decrease in size as necessary to achieve both within-class generalization and out-of-class generalization without requiring the use of non-target patterns in the training set. The resulting network learns compact closed decision boundaries when trained with target data only. We show results of applying the network to the Iris data set (Fisher (1936), Annals of Eugenics, 7(2), 179–188). Advantages of this approach include its inherent ability for one-class generalization, freedom from characterizing the non-target class, and the ability to form closed decision boundaries for multi-modal classes that are more complex than hyperspheres without requiring inversion of large matrices.

Cueing, feature discovery, and one-class learning for synthetic aperture radar automatic target recognition

Abstract The exquisite capabilities of biological neural systems for recognizing target patterns subject to large variations have motivated us to investigate neurophysiologically-inspired techniques for automatic target recognition. This paper describes a modular multi-stage architecture for focus-of-attention cueing, feature discovery and extraction, and one-class pattern learning and identification in synthetic aperture radar imagery. To prescreen massive amounts of image data, we apply a focus-of-attention algorithm using data skewness to extract man-made objects from natural clutter regions. We apply self-organizing feature discovery algorithms that uniquely characterize targets in a reduced dimension space and use self-organizing one-class classifiers for learning target variations. We also develop a distance metric for partial obscuration recognition. We present performance results using simulated SAR data and test for within-class generalization using nontrained targets including both in-the-clear and partially obscured examples. We test for between-class generalization using non-trained targets including both in-the-clear and partially obscured examples. We test for between-class generalization using near-target data.

Robot control systems: A survey

Abstract Robot manipulators have attracted considerable interest from researchers both in universities and industry during recent years. This interest covers a broad spectrum from task planning, robot language and artificial intelligence to mechanics, sensing and control. This survey paper addresses the area of robot position control and attempts to give an overview of the basic problems involved and some existing solutions.

Feature discovery via neural networks for object recognition in SAR imagery

A two-stage self-organizing neural network architecture has been applied to object recognition in synthetic aperture radar imagery. The first stage performs feature extraction and implements a two-layer neocognitron. The resulting feature vectors are presented to the second stage, an ART 2-A classifier network, which clusters the features into multiple target categories. Training is performed off-line in two steps. First, the neocognitron self-organizes in response to repeated presentations of an object to recognize. During this training process, discovered features and the mechanisms for their extraction are captured in the excitatory weight patterns. In the second step, neocognitron learning is inhibited and the ART 2-A classifier forms categories in response to the feature vectors generated by additional presentations of the object to recognize. Finally, all training is inhibited and the system tested against a variety of objects and background clutter. The results of the initial experiments are reported.<<ETX>>

Multispectral rock-type separation and classification

This paper shows the possibility of separating and classifying remotely-sensed multispectral data from rocks and minerals onto seven geological rock-type groups. These groups are extracted from the general categories of metamorphic, igneous and sedimentary rocks. This study is performed under ideal conditions for which the data is generated according to laboratory hyperspectral data for the members, which are, in turn, passed trough the Multispectral Thermal Imager (MTI) filters yielding 15 bands. The main challenge in separability is the small size of the training data sets, which initially did not permit the reliable estimation of the second-order statistics for every class. To enable Bayesian classification, the original training data is linearly perturbed with the addition of minerals, vegetation, soil, water and other valid impurities. As a result, the size of the training data is significantly increased and estimates of the covariance matrices are obtained. An eigenvalue analysis is used to generate a set of reduced (five) multispectral vectors, viz., feature vectors, providing principal information about the data. In addition, a nonlinear band-selection method is also employed, based on spectral indices, comprising a small subset of all possible ratios between bands. By applying three optimization strategies, optimal combinations of two and three ratios are found that provide reliable separability and classification between all seven groups. To set a benchmark to which the MTI capability in rock classification can be compared, an optimization strategy is performed for the selection of optimal multispectral filters, other than the MTI filters, and an improvement in classification is predicted when these filters are used.

Feature discovery in gray level imagery for one-class object recognition

Feature extraction transforms an objects image representation to an alternate reduced representation. Feature selection can be time-consuming and difficult to optimize so we have investigated unsupervised neural networks for feature discovery. We first discuss an inherent limitation in competitive type neural networks for discovering features in gray level images. We then show how Sangers Generalized Hebbian Algorithm (GHA) removes this limitation and describe a novel GHA application for learning object features that discriminate the object from clutter. Using a specific example, we show how these features are better at distinguishing the target object from other nontarget objects with Carpenters ART 2-A as the pattern classifier.<<ETX>>

Sensor-driven, fault-tolerant control of a maintenance robot

A robot system has been designed to do routine maintenance tasks on the Sandia Pulsed Reactor (SPR). The use of this Remote Maintenance Robot (RMR) is expected to significantly reduce the occupational radiation exposure of the reactor operators. Reactor safety was a key issue in the design of the robot maintenance system. Using sensors to detect error conditions and intelligent control to recover from the errors, the RMR is capable of responding to error conditions without creating a hazard. This paper describes the design and implementation of a sensor-driven, fault-tolerant control for the RMR. Recovery from errors is not automatic; it does rely on operator assistance. However, a key feature of the error recovery procedure is that the operator is allowed to reenter the programmed operation after the error has been corrected. The recovery procedure guarantees that the moving components of the system will not collide with the reactor during recovery.

Anomaly detection using simulated MTI data cubes derived form HYDICE data

In this work we quantify the separability between specific materials and the natural background by applying receiver operating curve (ROC) analysis to the residual errors from a linear unmixing. We apply the ROC analysis to quantify performance of the multi-spectral thermal imager (MTI). We describe the MTI imager and simulate its data by filtering HYDICE hyperspectral imagery both spatially and spectrally and by introducing atmospheric effects corresponding to the MIT satellite altitude. We compare and contrast the individual effects on performance of spectral resolution, spatial resolution, atmospheric corrections, and varying atmospheric conditions.

Feature discovery on segmented objects in SAR imagery using self-organizing neural networks

In this paper we investigate the applicability of the feature extraction mechanisms found in the neurophysiology of mammals to the problem of object recognition in synthetic aperture radar imagery. Our approach presents multiple views of target objects to a two-stage-organizing neural network architecture. The first stage, a Neocognitron, performs two layers of feature extraction. The resulting feature vectors are presented to the second stage, an ART-2A classifier self-organizing neural network which clusters the features into multiple object categories. In our first experiments reported in a previous paper, the Neocognitron was trained on raw SAR imagery. The architecture was able to recognize a simulated vehicle at arbitrary azimuthal orientations at a single depression angle while rejecting clutter as well as other vehicles. Feature extraction on raw imagery yielded features that were robust but difficult to interpret. We have performed new experiments in which the self-organization process is used to discover features separately in shadow and bright returns from objects to be recognized. feature extraction on shadow returns yields oriented contrast edge operators suggestive of bipartite simple cells observed in the striate cortex of mammals. Feature extraction on the specularity patterns in bright returns yield a mixture of orientation-independent operators similar to those found in the retina, and a collection of symmetric oriented contrast edge operators. These operators are formed at multiple positions within the receptive fields during the self-organization process and collectively resemble a two-dimensional Haar basis set. we merge the feature operators discovered separately in shadow and bright returns into a combined feature extractor front end. This front end is designed to extract the desired features from raw imagery. We compare the performance of the earlier two-stage neural network with a modified network using the new feature set.

Road segmentation using multipass single-pol synthetic aperture radar imagery

Synthetic aperture radar (SAR) is a remote sensing technology that can truly operate 24/7. Its an all-weather system that can operate at any time except in the most extreme conditions. By making multiple passes over a wide area, a SAR can provide surveillance over a long time period. For high level processing it is convenient to segment and classify the SAR images into objects that identify various terrains and man-made structures that we call “static features.” In this paper we concentrate on automatic road segmentation. This not only serves as a surrogate for finding other static features, but road detection in of itself is important for aligning SAR images with other data sources. In this paper we introduce a novel SAR image product that captures how different regions decorrelate at different rates. We also show how a modified Kolmogorov-Smirnov test can be used to model the static features even when the independent observation assumption is violated.