Abstract: In this paper, Sacramento, California-San Joaquin River Delta was taken as the study area, and HyMap hyperspectral data with 3 m spatial resolution acquired in June 2007 combined with ground truth data were used for pattern recognition of floating-leaved vegetation in the study area.The study was based on the spectral differences of wetland vegetations, and the "trilateral" parameters of vegetation were analyzed.Then the authors selected suitable vegetation indices combined with "trilateral" parameter features and built a decision tree model to extract the floating-leaved vegetation of the study area in comparison with the maximum likelihood classification results.The results show that the use of decision tree classification model can achieve overall accuracy of 82.68%, and that, compared with the maximum likelihood method, the total accuracy was improved by 6%, which can well identify the floating-leaved vegetation in the wetland vegetation of the study area.