Artificial Intelligence and Machine Learning in Optical Information Processing: introduction to the feature issue.

This special feature issue covers the intersection of topical areas in artificial intelligence (AI)/machine learning (ML) and optics. The papers broadly span the current state-of-the-art advances in areas including image recognition, signal and image processing, machine inspection/vision and automotive as well as areas of traditional optical sensing, interferometry and imaging.

Practical data mining and machine learning for optics applications: introduction to the feature issue.

Data mining algorithms utilize search techniques to explore hidden patterns and correlations in the data, which otherwise require a tremendous amount of human time to explore. This feature issue explores the use of such techniques to help understand the data, build better simulators, explain outlier behavior, and build better predictive models. We hope that this issue will spur discussions and expose a set of tools that can be useful to the optics community.

What can we learn from the shape of a correlation peak for position estimation?

Matched filtering is a robust technique to identify and locate objects in the presence of noise. Traditionally, the amplitude of the correlation peak is used for detection of a match. However, when distinguishing objects that are not significantly different or detecting objects under high noise imaging conditions, the normalized peak amplitude alone may not provide sufficient discrimination. In this paper, we demonstrate that measurements derived from the shape of the correlation peak offer not only higher levels of discrimination but also accurate position estimation. To our knowledge, this is the first time such features have been used in a real-time system, like the National Ignition Facility, where such techniques enable real-time, accurate position estimation and alignment under challenging imaging conditions. It is envisioned that systems utilizing matched filtering will greatly benefit from incorporating additional shape based information.

Artificial neural networks using complex numbers and phase encoded weights.

The model of a simple perceptron using phase-encoded inputs and complex-valued weights is proposed. The aggregation function, activation function, and learning rule for the proposed neuron are derived and applied to Boolean logic functions and simple computer vision tasks. The complex-valued neuron (CVN) is shown to be superior to traditional perceptrons. An improvement of 135% over the theoretical maximum of 104 linearly separable problems (of three variables) solvable by conventional perceptrons is achieved without additional logic, neuron stages, or higher order terms such as those required in polynomial logic gates. The application of CVN in distortion invariant character recognition and image segmentation is demonstrated. Implementation details are discussed, and the CVN is shown to be very attractive for optical implementation since optical computations are naturally complex. The cost of the CVN is less in all cases than the traditional neuron when implemented optically. Therefore, all the benefits of the CVN can be obtained without additional cost. However, on those implementations dependent on standard serial computers, CVN will be more cost effective only in those applications where its increased power can offset the requirement for additional neurons.

Convergence in optical and digital pattern recognition: introduction to the feature issue.

This Applied Optics feature issue, in addition to optical and digital pattern recognition, includes their convergence in the form of computational imaging.

Hardware accelerated optical alignment of lasers using beam-specific matched filters.

Accurate automated alignment of laser beams in the National Ignition Facility (NIF) is essential for achieving extreme temperature and pressure required for inertial confinement fusion. The alignment achieved by the integrated control systems relies on algorithms processing video images to determine the position of the laser beam images in real time. Alignment images that exhibit wide variations in beam quality require a matched-filter algorithm for position detection. One challenge in designing a matched-filter-based algorithm is to construct a filter template that is resilient to variations in imaging conditions while guaranteeing accurate position determination. A second challenge is to process images for thousands of templates in under a second, as may be required in future high-energy laser systems. This paper describes the development of a new analytical template that captures key recurring features present in the beam image to accurately estimate the beam position under good image quality conditions. Depending on the features present in a particular beam, the analytical template allows us to create a highly tailored template containing only those selected features. The second objective is achieved by exploiting the parallelism inherent in the algorithm to accelerate processing using parallel hardware that provides significant performance improvement over conventional processors. In particular, a Xilinx Virtex II Pro field programmable gate array (FPGA) hardware implementation processing 32 templates provided a speed increase of about 253 times over an optimized software implementation running on a 2.2 GHz AMD Opteron core.

Detection and tracking of the backreflection of potassium dihydrogen phosphate images in the presence or absence of a phase mask.

The potassium dihydrogen phosphate (KDP) crystals present in the final optics assembly at the National Ignition Facility (NIF) are used for conversion of an infrared laser light beam into an ultraviolet beam. The conversion is highest for a certain incident angle, the alignment of which is determined from the position of the backreflection beam, which exhibits a distinct characteristic shape. When a phase-plate device is introduced before the final assembly to increase the uniformity of the beam, the backreflection pattern changes drastically. The algorithm that is best for tracking the special-shaped beam is no longer suitable for tracking the phase-modified beam. We discuss our detection schemes for both situations. In particular, we demonstrate how the algorithm senses the modified beam by using a newly proposed criterion of correlation peak pedestal area and executes an alternate algorithm in real time without operator intervention. This new algorithm continuously tracks the beam pattern to guarantee reliable and repeatable sensing. Results from simulation and real-world implementation of the algorithm at the NIF facility are presented.

Optimal position estimation for the automatic alignment of a high-energy laser.

The alignment of high-energy laser beams for potential fusion experiments demands high precision and accuracy by the underlying positioning algorithms whether it be for actuator control or for monitoring the beam line for potential anomalies. The feasibility of employing on-line optimal position estimators in the form of model-based processors to achieve the desired results is examined. We discuss the modeling, the development, the implementation, and the processing of model-based processors applied to both simulated and actual beamline data.

Sub-imaging technique to improve phase only filter search capability.

A novel application of a phase only filter model, which is implementable in the optical domain, is proposed. In this application, automated target tracking is accomplished with a novel sub-imaging technique with correlation tracking inside a radius of interest. In this technique the image is subdivided and the correlation is tracked by comparing only the autocorrelation of the filter with itself. A radius of interest is used to reduce the number of computations required to track a moving target. Real-world video images are used to demonstrate the performance of the model.