- author = "Jill Sachie Kobashigawa Nakatsu",
- title = "Genetic programming applications in electromagnetics",
- school = "Electrical Engineering, University of Hawaii at Manoa",
- year = "2012",
- address = "USA",
- month = May,
- keywords = "genetic algorithms, genetic programming, electromagnetics, UXO, ANN",
-
URL = "
https://scholarspace.manoa.hawaii.edu/items/60eefc54-c6ce-4d02-9746-308176dc5d9a",
-
URL = "http://hdl.handle.net/10125/101315",
-
URL = "https://scholarspace.manoa.hawaii.edu/server/api/core/bitstreams/19b2b647-f256-44fb-91a9-c5d8d12e7734/content",
- size = "78 pages",
-
abstract = "This thesis considers the application of Genetic
Programming (GP) to create computer programs that can
solve both classification and metamaterial design
problems in the area of electromagnetics (EM).
Specifically, GP is used to develop an automatic target
classification algorithm and is combined with the
patterning Lindenmayer system for the development of a
metamaterial design program. For both studies, GP is
compared to other popular artificial intelligence (AI)
techniques in each area such as the Neural Networks
(NN) and the Genetic Algorithm (GA) methods. It is
shown that Genetic Programming provides improved
classification results and when applied to design work
leads to unconventional and global optimal
solutions.
In the target classification of buried objects it is desired to develop an accurate and reliable analysis and classification of electromagnetic data for buried unexploded ordnance (UXO) discrimination. The classification of this data is vital to not only clear buried UXO leftover from war and military training areas across the world with minimal false alarm rates but also to provide opportunities to use this land for housing and business development. GP is compared with neural networks, a popular classification technique, for the classification of UXO scattering patterns. Three classification scenarios with various levels of difficulty were examined and in all cases GP outperformed the NNs.
For the metamaterial design study, a GP program was developed that generates novel, efficient, and unintuitive 'broadband' metamaterial designs. There has been no established methodology for developing a successful design of ultra wideband and low frequency metamaterial structures and to this end; GP is used to investigate the development of unconventional designs. A metamaterial design system combining GP with Lindenmayer system (L-system) patterning rules was developed and used. A Matlab toolbox which controls both the GP algorithm and the full EM wave simulation in HFSS was also developed and utilized in the comparison of the GP-L system to the genetic algorithm. It is shown that GP is indeed capable of developing designs with improved performance from those reported using the GA methods. This thesis includes a detailed description of the developed GP code, fitness function, and obtained results from both studies.",
