oksi

HYPERSPECTRUM NEWS LETTER
Vol 2, No. 1, MAY 1995

The above AVIRIS data-cube was taken from NASA/JPL FTP site

CONTENTS

HELLO: The Internet 101
Why do I like e-mail as a means of communicating? There may be many answers; here are a few: a) as compared with the telephone, it is non disruptive, I read my mail and respond at my convenience, b) I don't have to give off-the-cuff responses, c) I don't get busy signals when dialing out, nor am I placed on hold, d) I do not have to navigate the voice mail choices [have you ever selected choices from a never-ending menu only to get to a dead-end (and then you can't backstep)?] e) I can compose my mail at a time most convenient to me, review, edit and send a polished message, with a copy for my records, f) it's faster than regular mail, it is as fast as a fax, g) I get immediate notification (undelivered returned mail) if I used the wrong address, h) an e-mail message can be printed and is more easily understood than a lengthy voice mail. On the down side, one loses the personal contact, but the telephone is always there to pick up and make that call when necessary.

The difficulties that some may have with e-mail stem from an "old" technology. With present Windows-based (and Mac) Internet communications software, it is as easy as a click of the mouse. No longer is there a need to memorize UNIX command-line syntax. What's more, one can send and receive files, graphics, text, video-clips, sound-bytes, etc.

The obvious next step is beyond e-mail. The amount of information accessible via the Internet is growing at an explosive rate. The introduction of Mosaic and its various successors makes surfing the net a breeze (even the Chairman of the Board can do it now). I truly believe that anyone who does not stay on top of advances in communications technologies, will be at a major disadvantage. As an indication of things to come, consider the following. NASA and some other agencies have already stopped distributing a hard copy of the SBIR/STTR solicitations. These are available only electronically via a BBS, FTP or WWW site. One major scientific organization has recently held an international conference over the Internet. Papers were posted and discussed over a period of time through a listserver. More and more organizations and universities post public domain technical papers (electronic reprints) and specialized software for retrieval via anonymous FTP (file transfer protocol). An increasing number of technical organizations also post abstracts of papers on WWW (world wide web) servers (also see the footnote on this page).

Yellow and white pages services, as well as other search tools are available, and easy to use, with the recent crop of Windows/Mac based communications software. And the best news is that many of these excellent programs are free (yes, you guessed it, via Anonymous FTP). So, happy :-) Internet surfing.

Nahum Gat, Editor


Report Review
Military Utility of Multispectral and Hyperspectral Sensors. A Report Published by IRIA, Nov. 1994; includes eight Chapters, two Appendices, bibliography in each chapter. Contents: Overview, Background, Phenomenology, Multispectral and Hyperspectral Sensing, Modeling and Simulation, Spectral Algorithms, Potential Defense Applications, Examples of Products for Defense Applications, Material Phenomenology, and Airborne Hyperspectral Systems.

If you have been involved in the field, working on sensors, and algorithms, faithfully reading papers and conference proceedings, for the past few years, then you'll probably find nothing new. But most of us have not, so that this state-of-the-art report is an excellent summary. It is also an excellent introduction to the topic for those with sufficient background in spectroscopy, remote sensing, electro-optical sensors, etc. What's not included is sufficient information about commercial hyperspectral analysis software packages. The inclusion of more technical design details of the many sensors discussed in the report would have been more useful; the engineers among the readers would have appreciated seeing optics layouts, data handling issues, etc. Also omitted are other applications of multi and hyperspectral sensors. This is understandable, though, since the report is concerned with military and utility. Nevertheless, as I have been advocating, the similarities among the various fields of applications make this report of interest even to those in the medical field (as long as an open mind is kept).


Company Profile
"Why do you publish this newsletter?" I'm often being asked, with a second trailing question whether I'm a freelance writer. The answers are simple. Several years ago I became involved in, and found the broad topic of hyperspectral systems to be fascinating (see OKSI profile below). The newsletter is simply a way of communicating with a rapidly increasing community in this field. I have no intentions of becoming a freelance writer or a magazine publisher. The question prompted me, however, to include a new feature. I'd like to use this opportunity to invite others to feature a profile of their company, department, or whatever, related to involvement in hyperspectral systems.

Hyperspectrum at OKSI:
As a small business, OKSI adapted several novel operational paradigms. One is the virtual office and telecommuting: there are certain aspects of our work that can be done anywhere. A physicist can develop as good a phenomenology math model working in the office, at home, or on the beach in Hawaii. The second paradigm, is augmenting our man-power using retirees as part time employees. Such an arrangement works exceptionally well for both sides. The retirees typically do not want a full time job yet they want to stay in touch with technology, engineering and science, and to contribute from their vast experience. Combining telecommuting with part time highly skilled employees gives OKSI a mix of technical skills seldom found in a small company. The non-virtual aspect is OKSI's lab in which real hardware is being designed, assembled, and tested.

Here is a brief review of some activities in the field of hyperspectrum at OKSI:

The Intelligent Missile Seeker (IMS) project was purposed to demonstrate the use of hyperspectral techniques for real-time discrimination between airborne targets and countermeasures. For collecting signatures, a sensor was built comprising a visible to near infrared (VNIR), and a midwave infrared (MWIR) imaging spectrometers using a common all-reflective aperture. With a grating for dispersion, the sensors use a 2562 CCD and 160x120 InSb FPAs, respectively. Data are digitized in the sensor and passed through an RS-422 to a frame grabber in a PC. Target/ countermeasure discrimination is accomplished in a transform domain using hyperspectral math transforms, a technique specifically developed for real-time hyperspectral applications.

The Thermal Infrared Imaging Spectrometer (TIRIS) is a unique, first of its kind, staring FPA, grating-based imaging spectrometer that operates between 7.5 to 14.0 śm. The 64x20 Si:As FPA is cooled to 10K, but the optical train is uncooled. A prototype is currently undergoing laboratory characterization tests, and will shortly be deployed in the field.

One application of the TIRIS is the passive detection of pollutants in the atmosphere, soil, or water. Another OKSI program, therefore, is focused on the development of algorithms for real-time detection.

In another area, OKSI is developing neural network algorithms for hyperspectral data processing. The focus is on the application of NN training techniques that are greatly faster than current training methods used with common NN architectures. Specific applications include medical imaging, as well as remote sensing.

I'm afraid that's all I'm going to tell you folks. IMS and TIRIS sensor photos can be found at http://www.techexpo.com/opto-knowledge/.


Multispectral, Hyperspectral, or Ultraspectral -- Thought on Bandwidth.
When I started using computers we used to carry our code in a shoe box filled with punched cards, or in a tightly wrapped punched paper tape that would read into a teletype terminal. Then in the late 1970's as the home computer industry emerged I made an investment in a Commodore 64. It had a total memory of a whopping 64KB of RAM, a single floppy, but no hard drive. My first word-processor was the "Paperclip," which came on 2 5-1/4" 180KB floppies (which was a major advancement compared to the 8" -- frisbee-size -- floppies of the PDP11/34). Once Paperclip and my document were loaded into RAM I could cut, paste, move text around and do quite a few of the basic manipulations and document formatting I do using my present word-processor. There was a little problem, because of obvious memory limitations, occasionally I would have to swap the program and document floppies in the drive, but that was a mere inconveniency. Then in the early 1980's I switched to a PC that came with a giant 10MB hard-drive. No more swapping diskettes, applications and documents were all loaded directly from the hard drive. The early version of WordStar, and later-on Manuscript (the scientific word-processor) used up less than 1 MB. In addition, the hard drive was able to store a spreadsheet, a graphics program, various utilities, games, DOS, and of course all my documents. The conventional wisdom was that I'll never outgrow my 10 MB hard drive. Right.

Today, 10 MB isn't enough to install a word-processor alone. Yes, I certainly could not produce the Hyperspectrum newsletter in its present format using Paperclip, but there is a lesson there. Given the availability of GB-size hard drives, and inexpensive RAM chips, there isn't much incentive for the programmer to emphasize efficient resource utilization when developing a new application; and programmers do become sloppy. Knowing that they had only 64KB of RAM which had to be shared with the document, and no hard drive, swap files, or virtual memory, the Paperclip programmers must have been sweating hard to cram in all the goodies. Then there is greed, the users want a word-processor with all the bells and whistles, although most of us are only familiar with a fraction of what is built into a present day word processor, and use even less.

What we have here is a technology push (or a solution looking for a problem), on one hand, and a user willing to pay the price of a general purpose, non-specific, all-in-one package (the price is resource utilization: disk space, memory, speed, etc.). In this example the stakes are not great. But is a document produced using MS Word rev. 6.0, of a higher quality than the one produced using Paperclip (or for that matter one typed on an old fashion Selectric I)?

So when Landsat was equipped with the Multi-Spectral Scanner (MSS) with four bands, trying to demonstrate that multispectral sensing from space can be practical for resource management application, the designers must have been sweating hard to select each band. When the Thematic Mapper (TM) was added with seven bands, that must have felt like a 10MB hard drive. And most of the remote sensing work to this day is done with these two instruments. But now what? Hyperspectrum time. AVIRIS gives us 244 bands; HYDICE, 210; and when it is launched on the Lewis spacecraft TRWIS will provide 384 bands. I was recently introduced to the term "ultraspectral." A glance at the ultraspectral data revealed that it is conventional spectroscopy performed under open-air-path remote conditions. High resolution spectroscopy is of course not new. It has been used for material identification in the laboratory for over 100 years. What is new is the combination of imagery with spectroscopy, and perhaps, one may argue, the application of spectroscopy in remote sensing is also somewhat new. But the question is: "When are 250 bands 25 times better than 10 bands?" The stakes in this case are higher than with word-processors. Is this a case of hype and a concept being oversold?

Some of the stakes introduce real challenges in terms of present day technology. Hyperspectral systems collect large volumes of data in a short time. Issues include (i) data storage volume, (ii) data storage rate, (iii) downlink or transmission bandwidth, (iv) real-time A/D bandwidth and resolution, (v) computing bottle neck in data analysis, (vi) new algorithms for data utilization, and more. And even if the technological challenges are resolved, the question posted above still remains actual.

I'd dare to say that of all the possible users out there, nobody needs, wants, or knows what to do with all the data. The prevailing attitude today is that hyperspectral systems are going to solve many problems in detection, identification, etc. I'd like to suggest that when the data are collected a lot of confusion will emerge, not necessarily bringing the user closer to a solution to the problem of interest. Furthermore, I contend, that once a phenomenology is thoroughly analyzed, many of the problems will be solved using only a few spectral bands (I'd not be surprised if between 2 to 8 channels will solve most problems posed today).

Another question, of course is, "what is the minimum number of bands and which ones are required for a particular problem?" Are we going to answer this question using the arsenals of operational hyperspectral sensors? Unfortunately, the answer is not straight forward, for a number of reasons. First, beyond statistical techniques, we do not yet have the algorithmic optimization tools (other than to experimentally attempt every possible permutation). Second, the solution must be strongly coupled with the physics and the phenomenology. Bands (i.e., band center and band width), as in the case of the MSS/TM, should (and can) be "hand-picked" even before we build an instrument. The operational sensors' bands (center line and width) may not always provide the clue. In some cases bands may have to be binned, others may have to be split, or discarded all together.

So, will hyperspectral techniques produce higher quality information than multispectral? My answer to this question is the same as to the question whether MS Word 6.0 produces better documents than Paperclip does. If we are smart enough, however, we may learn from these systems how to chose our bands.


Tidbits & Interesting Reading
In case you have not noticed yet: For your convenience: HYPERSPECTRUM is now on the Internet. You can view and download the newsletter in HTML at: http://www.techexpo.com/opto-knowledge/hyperspectrum/ While visiting the site, view a comprehensive schedule of technical conferences, or lists of technical societies, technical magazines, high-tech companies free-listed, at: http://www.techexpo.com/
Attending a Conference in which Hyperspectrum related papers are presented? While sitting at the session, and your impressions are fresh, please jot down a few comments, and send us a brief overview of the papers. It will go into the next issue, and will immediately be posted on the Internet WWW edition of Hyperspectrum. You'll receive the appreciation of the whole hyperspectrum community.
Meeting Your Interests -- Multi & Hyperspectral Systems: An agenda for the frequent flyer. Conferences, and call for papers, related to multi- and hyperspectral techniques.
A COMPREHENSIVE SCHEDULE OF CONFERENCES, TECHNICAL SOCIETIES, & TECHNICAL MAGAZINES CAN BE FOUND ON THE INTERNET AT http://www.techexpo.com/tech_inf.html
A MV (MUST-VISIT) SITE FOR ENGINEERS, SCIENTISTS, MANAGERS, AND HI-TECH GEEKS.

Don't be old fashioned. If you organize a conference -- consider including a session on hyperspectral and multispectral imaging. Everyone else does. Then let us know if you want it listed here.

TO POST YOUR MEETING: COMPLETE AND SUBMIT INFORMATION TO: TechExpo .


OPTO-KNOWLEDGE SYSTEMS, INC.
1737 Third Street
Manhattan Beach, California 90266-6308
310/372-6665 FAX# 310/379-9842 E-mail: oksi@cerfnet.com
A CALL FOR INFORMATION: Join our staff of reporters and tipsters around the country. Send in any news worthy of printing. Use e-mail, snail-mail, fax, pony express. Just don't say we haven't given you enough options.
Prefer to receive this publication via the internet, tell us, so we don't waste paper mailing you the hardcopy! Send us the following e-mail, with your name and address as it appears on the label.
STOP CLUTTERING MY DESK! PLEASE REMOVE MY NAME FROM YOUR MAILING LIST. I'LL GET IT VIA THE INTERNET.
Return to the Hyperspectrum Index.
Return to Opto-Knowledge Homepage.
Browse TechExpo stuff.

Copyright: © You MAY copy this newsletter to your computer or print it for your own entertainment or amusement, or even for business and technical purposes. You MAY NOT mirror or redistribute the newsletter on another WWW site. You can, however, include a link to the newsletter on this site in your own documents.