William W. Stoner

Optical implementation of a feature-based neural network with application to automatic target recognition

An optical neural network based on the neocognitron paradigm [IEEE Trans. Syst. Man Cybern. SMC-13, 826-834 (1983)] is introduced. A novel aspect of the architectural design is shift-invariant multichannel Fourier optical correlation within each processing layer. Multilayer processing is achieved by feeding back the output of the feature correlator iteratively to the input spatial light modulator and by updating the Fourier filters. By training the neural net with characteristic features extracted from the target images, successful pattern recognition with intraclass fault tolerance and interclass discrimination is achieved. A detailed system description is provided. Experimental demonstrations of a two-layer neural network for space-object discrimination is also presented.

Retinal model with adaptive contrast sensitivity and resolution

A computer-simulated retina (called IRIS) is described which has useful features for computer vision.: IRIS discriminates small differences in reflected light when these differences occur in a restricted domain of space and time, and maintains sensitivity to these differences for a wide range of lighting environments. As prevailing light levels decrease and photon noise becomes significant, IRIS reduces its spatiotemporal resolution to provide greater redundancy. Electrically, IRIS could be implemented by 2-D lattice of photosensors attached to a passively conducting grid. Inthe present study, the model is implemented in FORTRAN. Similarities to the human retina are noted; the photosensor design is based on a photoreceptor model, and each conducting grid may be realized by tight-junction coupling of receptors and by horizontal-cell interconnections (effectively forming a syncytium).

Design and Optimization Electron Guns and Depressed Collectors Using the MICHELLE code in the ANALYST Environment

Electron gun and multistage depressed collector design remain significant tasks in new vacuum electron device development. The ability to accurately predict device performance continues to improve. Recent work has been done on improving the ability for modeling and design simulation environments to aid the designer in finding optimum configurations. As the simulation tools have improved to enable first- pass design success in some cases, the potential benefits of optimization techniques become even more significant. This paper discusses various procedures for optimization of electron guns as well as multistage depressed collectors.

Optoelectronically implemented neural network with a wavelet preprocessor

An optoelectronic neural network based upon the Neocognitron paradigm has been implemented at JPL and successfully demonstrated for automatic target recognition for both focal plane array imageries and range-Doppler radar signatures. A novel feature of this neural network architectural design is the use of a shift-invariant multichannel Fourier optical correlation as a building block for iterative multilayer processing. An innovative bipolar neural weights holographic synthesis technique was utilized to implement both the excitatory and inhibitory neural functions and dramatically increase its discrimination capability. In order to further increase the optoelectronic Neocognitrons self-organization processing ability, a wavelet preprocessor has been developed for feature extraction preprocessing (orientation, size, location, etc.). The addition of this wavelet processor would enable the neocognitron to dynamically focus on the incoming targets based on their known features and result in higher discrimination and lower false alarm rate. The theoretical analysis of an orientation and scale selective wavelet is provided. A multichannel optoelectronic wavelet processor using an e- beam complex-valued wavelet filter is also presented. Experimental demonstrations of wavelet preprocessing for feature extraction are also provided.

Pattern Recognition With A Neural Net

Following the neocognitron architecture described by Fukushima, an Artificial Neural System (ANS) has been programmed in Fortran and run on an IBM PC AT. Our independent experience with this ANS confirms the findings of Fukushima for the neocognitron architecture. Specifically we exercised both the learning and recognition modes of the ANS. In the learning mode, alphanumeric characters are learned and distinguished without instruction or outside correction of errors. In the recognition mode, alphanumeric characters are recognized with tolerance to position, scale and geometric distortion. We describe the neocognitron architecture and explain the basis of its operation for both the learning and recognition modes.

One-Dimensional To Two-Dimensional Transformations In Signal Correlation

We explore one-dimensional (1-D) to two-dimensional (2-D) transformations suitable for optical correlation of signals. An intuitive, geometrical development establishes and relates several 2-D formats including variations of the familiar falling raster format. Insight into the folded spectrum is gained by considering a matched spatial filter implementation of correlation between two signals in falling raster formats.

Signal processing approach to designing gradient index correctors for wide-field, concentric lenses

The signal processing method used in transaxial tomography, reconstruction from projections, is applied to the design of a spherical gradient index corrector for the concentric “noflare” lens. The method yields a spherical gradient index lens that corrects the entire field of the lens. Motivation for the method is provided by a comparison of the Schmidt and super-Schmidt camera designs.

Optimization of Electron Guns and Collectors using the 2D/3D Michelle and Anlayst Finite-Element Codes

This paper discusses various procedures for optimization of electron guns as well as multistage depressed collectors. The MICHELLE code has been interfaced to the ANALYST analysis package, which provides comprehensive support for finite-element electromagnetic analysis, including embedded computer-aided design (CAD) software, automated meshing, and both visual and numerical result processing.

Towards a statistical error estimate for convex-hull derived endmembers

The convex hull methods for estimating spectral endmembers are subject to bias errors: mixed pixel bias - if all of the available pixels are mosaics of all m endmembers, the convex-hull derived endmember spectra are biased towards the centroid of the true endmember spectra; noise bias - additive Gaussian measurement noise inflates the convex hull away from the centroid of the noise-free convex hull. The noise bias error grows with the pixel count. This vulnerability to mixed pixel bias and noise bias prompts the following questions. Does the convex hull method throw away information by discarding the pixels lying inside the convex hull? Can bias error estimates be developed for convex-hull derived endmembers? Can bias-resistant endmember estimation methods be found? What is the gain in accuracy of the endmember estimates with increasing pixel count? What is the gain in accuracy with increasing density of pixels in the n-dimensional neighborhood of the true endmember? The following analysis focuses on these questions by omitting all sources of noise and distortion except the number and distribution of the samples in the neighborhood of the endmember.

Optical implementation of a shift-invariant neocognitron

An optical neural network based upon the Neocognitron paradigm is introduced. A novel aspect of the architectural design is shift-invariant multichannel Fourier optical correlation within each processing layer. An innovative bipolar neural weights holographic synthesis technique is introduced to implement both the excitatory and inhibitory neural functions. Multilayer processing is achieved by iteratively feeding back the output of the feature correlator to the input spatial light modulator and updating the Fourier filters. By designing the neural net with characteristic features extracted from the target images, successful pattern recognition with intra-class fault tolerance and inter-class discrimination is achieved. A detailed system description is provided. Experimental demonstrations of a two-layer neural network for space objects discrimination is also presented.© (1993) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.