Rapid measurement of brown tide algae using Zernike moments and ensemble learning based on excitation-emission matrix fluorescence

Accurate real-time prediction of microalgae density has great practical significance for taking countermeasures before the advent of Harmful algal blooms (HABs), and the non-destructive and sensitive property of excitation -emission matrix fluorescence (EEMF) spectroscopy makes it applicable to online monitoring and control. In this study, an efficient image preprocessing algorithm based on Zernike moments (ZMs) was proposed to extract compelling features from EEM intensities images. The determination of the highest order of ZMs considered both reconstruction error and computational cost, then the optimal subset of preliminarily extracted 36 ZMs was screened via the BorutaShap algorithm. Aureococcus anophagefferens concentration prediction models were developed by combining BorutaShap and ensemble learning models (random forest (RF), gradient boosting decision tree (GBDT), and XGBoost). The experimental results show that BorutaShap_GBDT preserved the su-perior subset of ZMs, and the integration of BorutaShap_GBDT and XGBoost achieved the highest prediction accuracy. This research provides a new and promising strategy for rapidly measuring microalgae cell density.