Project Summary

Proposal Number:

Project Title:Novel Neural Network Technology for Very Fast Analysis of Hyperspectral Imagery

Small Business Concern:
Opto-Knowledge Systems, Inc. (OKSI)
1737 3rd Street
Manhattan Beach, CA 90266-6308

Research Institution:
Jet Propulsion Laboratory,
California Institute of Technology,
4800 Oak Grove Drive,
Pasadena, CA 91101

Principal Investigator/Project Manager: Nahum Gat, Ph.D.


Technical Abstract:
Existing image analyses techniques, for spaceborne multispectral imagery were developed based on the limited number of spectral bands provided by sensors such as the TM/MSS. These image interpretation methods are inadequate for processing data provided by hyperspectral sensors to be launched in the next few years, such as MODIS (36 non- contiguous bands) and TRWIS (384 bands). The OKSI/JPL team proposes to adapt and utilize novel advanced neural network paradigms and architectures for solving such very high data volume image interpretation problems. The proposed NN learning algorithms have been demonstrated to (i) be faster by orders of magnitude than conventional NN techniques, and (ii) guarantee global optimization with respect to all the NN parameters. These techniques apply to feedforward, static and dynamic recurrent, ART or hybrid network architectures, and can be used in supervised or unsupervised training, thus making the application of NN to large data volume hyperspectral image interpretation practical. Hyperspectral data will be used during Phase I for proof of principle demonstration and performance (speed and accuracy) benchmarking. The team will work to develop value-added data products based on EOS and other sensors' data.



Potential Commecial Applications:
Applications include all the traditional remote sensing areas but with much higher accuracy, resolution, and speed: (i) agriculture: crop stress management due to heat, drought, or storm damage; crop maturity analyses and harvesting time optimization in high water value crops; high spatial resolution fertilization and pesticide requirements determination, (ii) environmental monitoring including monitoring thermal pollution, superfund site cleanup assessment, oils spills, wetlands, etc., (iii) forestry and land management, and (iv) enhancement of the EOS science applications in terrestrial, ocean, atmosphere, snow and ice research. The use of the NN technology in the following areas is also described: Photodiagnosis, environmental monitoring, and military intelligence.