In this study, inversion models of phytoplankton pigment concentrations are built for the total chlorophyll a and seven diagnostic pigments (i.e., chlorophyll b, fucoxanthin, peridinin, 19'-hexanoyloxyfucoxanthin, 19'-butanoyloxyfucoxanthin, alloxanthin, and zeaxanthin). Specifically, given the field measured data of fluorescence excitation spectra, the feature representations of fluorescence excitation spectra are constructed, and the machine learning algorithm eXtreme Gradient Boosting (XGBoost) is employed to build these models. The validation indicates that the inversion models have good estimation accuracy, among which the inversion model of the total chlorophyll a has the highest accuracy (with the determination coefficient of 0.87, the mean absolute percentage error of 28.1%, and the root mean square error of 1.168 mg·m^-3). In addition, these pigment inversion models are applied to typical sections of the East China Sea, and vertical distribution features of pigment concentrations are obtained.