Alabama A&M University Research Institute is the research arm of an HBCU in Huntsville, Alabama. Two of the world’s most celebrated scientists (major international awards) work there – H, John Caulfield and R. Barry Johnson. Most of its work in domestic preparedness is in pattern recognition along with great and celebrated individuals in radar and electromagnetic wave sensors.

• It performs better than any technique we have found in terms of generalization (the ability to place new, un-trained-on objects in their proper es with high reliability.
• It gives wonderful results using very few examples in the training set – 10-20 as opposed to the thousands Learning Theory suggests are necessary for the same performance levels. We have a simple theoretical proof that it should do better than is possible best prior methods
• It is domain independent. It works on any kinds of measurements. So far, we have applied it to gamma ray spectra, colored images, spatial pattern recognition and location, texture discrimination, spectral processing: all with excellent results.
• It does best when we know what the interfering objects will be, but it also works quite well when the only available signals come from the to be recognized.
• It lends itself well to treatment with Boolean or Fuzzy logic.

• Discrimination and segmentation by user-defined groups (for example, we might have two groups both of which have blue and red elements)
• Fast and accurate target discrimination and location in hyperspectral image (HSI) cubes.
• If built into a specialized system, it is
  • Much more sensitive than HSI
  • Much cheaper than HSI
  • Much smaller and more tugged than HSI
  • Equally useful in target discrimination with HIS

There are two things to be gotten from replacing some conventional pattern recognition system with our Margin Setting algorithm that achieves spectacular discrimination of new, un-trained-on data.

1. You can achieve your current performance level with cheaper, less exotic means or
2. You can use the current sensor and achieve much better performance.

The methods we have researched allow either, because it is provably superior to current methods such as the Support Vector Machine, Boosting and bagging, etc. Its superiority is not surprising as it made a performance-enhancing change in the usual “ground rules” for pattern recognition.

To date we have applied Margin Setting to

• Spectral discrimination and segmentation (our primary thrust, but one not described here because of word count limits.
• Spatial location and recognition of objects to provide pose and scale information about the object as well as itsentity and location
• Recognition by texture (used to extract small regions of something unexpected in a field dominated by material of another texture
• Recognition by polarization (Directions and degree of polarization)
• Gamma ray spectroscopy (Licensed to Innovative American technologies)
• Hyperspectral data cubes (processed very quickly with superb results)
• IR images to extract temperature despite unknown spectral emissivity patterns)

Alabama A&M University Research Institute is the research arm of an HBCU in Huntsville, Alabama. It has some of the most celebrated opticists in the world. The contact person, for example, was recently awarded the highest prize (The Gold Medal) os SPIE – the world’s largest optical society and was keynote speaker the last time the world optics society (ICO) met in St. Petersburg. He will give his fourth straight banquet talk when the Information Society has its meeting in Salt Lake City this year.

Contact Information:

H. John CaulfieldPO Box 313Normal, AL 35763John.Caulfield@cim.aamu.eduhttp://aamuri.aamu.edu