Electronic Military & Defense magazine was developed for engineers, program managers, project managers, and those involved in the design and development of electronic and electro-optic systems for military, defense, and aerospace applications.
Issue link: https://electronicsmilitarydefense.epubxp.com/i/546858
Moreover, many modulation schemes require extremely precise alignment of multiple optical paths, precise timing of rotating or moving components, and multiple sensors or apertures. Optomechanical design complexity is a hurdle to inexpensive, portable, and well-performing polarimetric instruments. Also, data coming from an instrument contains more information than data coming from a typical color or monochrome camera. Specifically, a passive system delivers four times as much information, and an active system delivers 16 times as much information. The data processing requirements also are demanding, especially for active (Mueller matrix) imaging polarimeters. For example, say we want to obtain Mueller matrix data with the following parameters: • 0.44 megapixel image size, 12 bit depth per pixel data (requires 2 bytes per pixel) • 30 frames per second (fps) • Single color/monochrome image Then, with the best spatio-temporal system we have designed to date, we need a system that has a 1.75 megapixel actual sensor size and an actual camera frame rate of 240 fps, resulting in a data processing rate of ~850 MB/s! This rate is right at the limit of today's fastest commercially avail- able frame grabber cards. The processing also includes fast Fourier transforms (FFTs), linear transformations, and filtering of three-dimensional data, which also must be accomplished at the 850 MB/s rate. Our current portable Mueller matrix polarimeter (Figure 3) only has a camera with an actual frame rate of 30 fps, resulting in an effective Mueller matrix image rate of 3.75 fps. Although fast for a Mueller matrix polarim- eter, this still is lacking in speed for a TTI application. Another significant deficit to achieving TTI using polarimet- ric imagers is the lack of statistically robust polarimetric (espe- cially Mueller matrix) datasets. Most datasets are either sparse in terms of material and number of measurements, propri- etary or sensitive/restricted in some way, and/or measured on a small portion of the polarized bidirectional reflectance distribution function (pBRDF) hemisphere. Drawing from the text recognition, imaging science, and machine learning disciplines, we realize that sparse and restricted datasets must be addressed for classification results to be meaningful (i.e., training datasets for TTI tasks must contain enough statistical power for training to generalize to discrimination in the field). Robust datasets are particularly important for remote sensing classification tasks since object movement and orientation are outside of our control. Our group recently designed and soon will be building a polarimetric scatterometer (Figure 6), which can measure the pBRDF over the entire hemisphere, in both transmission and reflection geometries. This instrument, along with the instrument in Figure 3, can be used to build up a large database of Mueller matrix measurements for discrimination algorithms to be used in training. In contrast, the hyperspectral community has large public datasets available, especially satellite data. Recent Developments Benefiting Polarimetric TTI Applications Recent advancements that have resulted in the imminent utilization of polarimetry for TTI include: • Theoretical improvements in polarimetric instrument bandwidth (speed, frames per second) • Theoretical breakthroughs in polarimetric system design using linear systems theory • Reduction in instrument complexity by using task- based approaches (compressive sensing) • Application of robust classification techniques from the mathematics, computer science, and machine-learning communities to polarimetric data • Computational improvements provided by COTS GPU/ CPU hardware Our group has contributed to all five areas, and we cur- rently are building one of the only portable active (Mueller matrix) remote sensing polarimeters in the world. The two most important developments for realizing polarimetric TTI are linear systems theory, as a basis for system design, and use of compressive sensing/task-based techniques for spe- cific TTI tasks. A major change in the field began between 2010 and 2011 — stemming primarily from work by Charles LaCasse and Andrey Alenin — with the application of linear systems/communications theory to the description and design of Stokes polarimeters. Alenin generalized this theory to all polarimetric instruments, and we currently are using the theory to create bandwidth-optimal active instrument designs. In a nutshell, linear systems theory allows a designer to directly trade off noise performance, bandwidth (spectral, temporal, or spatial), and polarization parameters of inter- est in a systematic way, rather than an ad hoc way. Other groups viii also have made improvements to bandwidth and noise performance using the linear systems description ix . The linear systems description allows an instrument designer to actively work in the tradeoff space in order to meet real- world systems requirements. The linear systems framework also can be used to describe active or passive instruments that only measure a subset of the available parameters. In many TTI tasks, a subset of polarimetric measurements can be taken in order to meet a classification performance specification. For a given Mueller matrix training dataset (fully "tagged" or specified for target Technology 10 Figure 6: Advanced Sensing Laboratory polarized scatterometer, able to mea- sure full pBRDFs over the visible wavelength range Electronic Military & Defense Annual Resource, 5th Edition