The coastal wetland is an ecological transition zone at the junction of sea and land, and is one of the most biodiverse ecosystems in nature. As an important part of wetland ecosystems, studying the eco-stoichiometric characteristics of carbon, nitrogen and phosphorus is an effective way to understand plant growth status and adaptation strategies. Taking Yancheng Coastal Wetland in Jiangsu Province as the research area, this paper collected five dominant wetland plant samples and canopy hyperspectral data, including Spartina alterifora, Phragmites australis, Imperata cylindrica, Tamarix chinensis, and Suaeda salsa, and conducted hyperspectral inversion research on the eco stoichiometric characteristics of carbon, nitrogen and phosphorus of plants. The results showed that the random forest (RF) model was the best inversion model for P. australis, I. cylindrica and T. chinensis, the partial least squares (PLSR) model was the best inversion of S. alterifora, and the model with the highest inversion accuracy of S. salsa was BP Neural Network (BPNN) model. This study shows that the use of hyperspectral data can achieve accurate inversion of the ecological stoichiometric characteristics of carbon, nitrogen and phosphorus in wetland plants. Different models have different inversions for different wetland plants. The RF model has the strongest inversion stability and is a better model for inversion of wetland plant ecological stoichiometric characteristics.