In this paper, Intelligent Optical Systems, Inc. reports on our progress in using neural network signal processing algorithms for the enhancement of sensor signals from a multigas optical sensor under development for NASA. We found that a 4x8x3 neural network yielded superior results over the last squares (LS), partial least squares (PLS), and principal components regression (PCR) algorithms in estimating oxygen, water vapor, and temperature.
This paper presents an overview of current ongoing research and design efforts conducted by Intelligent Optical Systems, Inc. in the area of hardware-based color segmentation. We discuss the specifics of the design of a microchip that combines a hardwired hybrid neural network with on-chip imaging. Several preliminary tests show high approximation ability of our scheme. The single-chip implement has many advantages. The final product will consists of an RGB pixel array with infinite color depth and a neural network capable of high speed image segmentation.
We have tested several predictive algorithms to determine their ability to learn from and find relationships between large numbers of variables. The purpose of this test is to produce control algorithms for sophisticated devices like particle accelerators. In particular we use COMFORT, a particle accelerator simulator, to generate large amounts of data. We then compared results among several fundamentally different types of algorithms, including least squares and hybrid neural networks. Our data indicate which algorithms perform the best on the basis of performance and training times.
Low-cost, compact, and robust color sensor that can operate in real-time under various environmental conditions can benefit many applications, including quality control, chemical sensing, food production, medical diagnostics, energy conservation, monitoring of hazardous waste, and recycling. Unfortunately, existing color sensor are either bulky and expensive or do not provide the required speed and accuracy. In this publication we describe the design of an accurate real-time color classification sensor, together with preprocessing and a subsequent neural network processor integrated on a single complementary metal oxide semiconductor (CMOS) integrated circuit. This one-chip sensor and information processor will be low in cost, robust, and mass-producible using standard commercial CMOS processes. The performance of the chip and the feasibility of its manufacturing is proven through computer simulations based on CMOS hardware parameters. Comparisons with competing methodologies show a significantly higher performance for our device.
An adaptive multilayer optical neural network with all-optical forward propagation including optical thresholding by a liquid crystal light valve (LCLV) is described. It has a large number of modifiable optical interconnections that are implemented by liquid crystal television screens, and it has a modular structure enabling the cascading of layers, each layer with its own light source. Sigmoid fits to response curves of four LCLVs are evaluated and their suitability as optical thresholding functions is examined on the basis of neural network simulations.
Neural networks are a primary candidate architecture for optical computing. One of the major problems in using
neural networks for optical computers is that the information holders: the interconnection strengths (or weights) are
normally real valued (continuous), whereas optics (light) is only capable of representing a few distinguishable intensity
levels (discrete). In this paper a weight discretization paradigm is presented for back(ward error) propagation
neural networks which can work with a very limited number of discretization levels. The number of interconnections
in a (fully connected) neural network grows quadratically with the number of neurons of the network. Optics can
handle a large number of interconnections because of the fact that light beams do not interfere with each other.
A vast amount of light beams can therefore be used per unit of area. However the number of different values one
can represent in a light beam is very limited. A flexible, portable (machine independent) neural network software
package which is capable of weight discretization, is presented. The development of the software and some experiments
have been done on personal computers. The major part of the testing, which requires a lot of computation,
has been done using a CRAY X-MP/24 super computer.