Study on fluorometric discrimination of phytoplankton based on time-series vectors of wavelet transform

The feasibility of using time domain of wavelet transform as characteristics to establish a fluorometric discrimination method of phytoplankton was discussed. Twelve phytoplankton species belonging to nine genera of five divisions were studied. Five steps were introduced: firstly, the feasibility of utilizing 3D fluorescence spectra (3D-FS) to discriminate phytoplankton was discussed; the relative standard deviation (RSD) and included angle cosine (IAC) were used as the test criterion. 3D-FS had such potentials, for most RSD were <5% and most IAC were >0.990. Secondly, the 3D-FS were decomposed by db7 wavelet and time-series vectors (TSVs) were generated. Thirdly, the optimal characteristic spectra (OCS) were selected from the TSV by Bayesian linear discriminant analysis (BLDA). The ability of OCS to classify phytoplankton was tested, and the correct classification ratios (CCRs) at different levels were obtained. Most CCRs were 90-100% at the species level. They were >98% at the genus level, and >99% at the division level. Fourthly, the growth and light stability of the OCS were tested. Both stabilities were high with lower RSD (<3%) and higher IAC (>0.999) compared with 3D-FS. Fifthly, a "database of reference spectra" consisting of 46 reference spectra was established by hierarchical cluster analysis (HCA). Based on this, the discrimination method of phytoplankton species was established by nonnegative least squares (NNLSs). Most reference spectra were representative to phytoplankton species; and had moderate anti-noise ability: With noise 98% at the genus level and >99% at the division level. 20% noise was a larger interference which made CDRs down to 85% at the genus level and to 99% at the division level. A fluorometric discrimination method of phytoplankton could be established based on TSV of wavelet transform.