A fluorescence discrimination technique for the dominant algae species developed by Wavelet packet
Yali Duan,Rongguo Su,Shuwei Xia,Shanshan Zhang,Cui Zhang,Xiulin Wang
DOI: https://doi.org/10.1109/RSETE.2011.5965870
2011-01-01
Abstract:In this study, a fluorescence spectra discrimination technique for red tide algae by Wavelet packet transform was developed. The fluorescence excitation-emission spectra were determined by fluorescence spectrophotometer for 24 red tide algae. Then the coefficient vectors (candidate feature spectra) were obtained by decomposition of Coiflets-2 wavelet packet. Bayesian discrimination was applied to select the feature spectra from the feature spectra and the norm spectra database was established by Cluster analysis. Finally, the discrimination technique was developed by multivariate linear regression and non-negative least squares. The results showed: for the simulative samples, when the dominant algae species accounted for 60%, 70%, 80% 90% of the gross biomass, the CDR (correct discrimination ratio) of the dominant algae species at division level were 83.5%, 99.1%, 99.6% and 99.9% with the average relative content of 58.52%, 68.36%, 77.66%, 86.33%, respectively; and when the dominance of the dominant algae species accounted for 60%, 70%, 80%, 90%, 100%, the CDR of the dominant algae species at genus level were 86.13%, 94.91%, 95.25%, 96.78%,97%, respectively. For 12 samples collected from the mesocosm experiments in Maidao Bay, the CDR of the dominant algae species was 91.7% at division level, and that of the dominant algae species was 80% at genus level for the five samples that the dominance of dominant algae species reached to 80% according to the results of microscopic counting. For 12 samples collected from Jiaozhou Bay in August 2007, the CDR of of the dominant algae species was 100% at division level, and the dominant algae species of two samples were correctly recognized at genus level for three samples which dominant algae species accounted for 80% of the gross biomass according to the results of microscopic counting.