Optimal classification of standoff bioaerosol measurements using evolutionary algorithms

Author
Nyhavn, Ragnhild
Moen, Hans Jonas Fossum
Farsund, Øystein
Rustad, Gunnar
Date Issued
2011
Keywords
Bioaerosoler
Målinger
Mønstergjenkjenning
Permalink
http://hdl.handle.net/20.500.12242/802
https://publications.ffi.no/123456789/802
DOI
10.1117/12.883919
Collection
Articles
Description
Nyhavn, Ragnhild; Moen, Hans Jonas Fossum; Farsund, Øystein; Rustad, Gunnar. Optimal classification of standoff bioaerosol measurements using evolutionary algorithms. Proceedings of SPIE, the International Society for Optical Engineering 2011 ;Volum 8018. s. -
870852.pdf
Size: 2M
Abstract
Early warning systems based on standoff detection of biological aerosols require real-time signal processing of a large quantity of high-dimensional data, challenging the systems efficiency in terms of both computational complexity and classification accuracy. Hence, optimal feature selection is essential in forming a stable and efficient classification system. This involves finding optimal signal processing parameters, characteristic spectral frequencies and other data transformations in large magnitude variable space, stating the need for an efficient and smart search algorithm. Evolutionary algorithms are population-based optimization methods inspired by Darwinian evolutionary theory. These methods focus on application of selection, mutation and recombination on a population of competing solutions and optimize this set by evolving the population of solutions for each generation. We have employed genetic algorithms in the search for optimal feature selection and signal processing parameters for classification of biological agents. The experimental data were achieved with a spectrally resolved lidar based on ultraviolet laser induced fluorescence, and included several releases of 5 common simulants. The genetic algorithm outperform benchmark methods involving analytic, sequential and random methods like support vector machines, Fisher's linear discriminant and principal component analysis, with significantly improved classification accuracy compared to the best classical method.
View Meta Data