Hyperspectral data is rich in information and bands, which can provide a more comprehensive basis for geophysical analysis, but at the same time, it also increases the complexity and interference of data analysis, especially in low signal-to-noise ratio applications such as remote sensing monitoring of water quality. Traditional band selection often uses correlation coefficient and other methods to select the identification band in many spectral bands and to analyze the data on the selected band set. In this paper, based on the constrained energy minimization (CEM), a target-oriented band selection algorithm is proposed, which is called CEM-based band selection (CBS). The signal matching filter is used to find the band with a high matching degree with the target vector from the observation vector, and then combined with the orthogonal principle to maximize the selection of a subset of bands that have a high degree of matching with the target vector and low redundancy of the band vector. Based on the determination of the components in the water quality monitoring, the hyperspectral data of the Liaohe estuary test area was collected and combined with the synchronous field water sample data to predict the nitrogen and phosphorus content in the Liaohe waters. Comparing the band selection results of the CBS algorithm with the band selection results of the Pearson correlation coefficient (PCC), the significant band subsets obtained by the two methods are used as variables to carry out stepwise regression analysis, and multiple regression models are established to further test the accuracy of the model and analyze the average relative error between the predicted value and the true value. In the accuracy test of the total phosphorus concentration model, the average relative error of the model obtained by the PCC algorithm is 20.7%, and the average relative error of the model obtained by the CBS algorithm is 8.17%. In the accuracy test of the total nitrogen concentration model, the average relative error of the model obtained by the PCC algorithm is 16.8%, and the average relative error of the model obtained by the CBS algorithm is 12.4%. The results of the data analysis show that the band subset obtained by the CBS algorithm is superior to the traditional selection method based correlation coefficient in the ability of nitrogen and phosphorus concentration inversion.