Chlorophyll-a, the main pigment in phytoplankton, is an indicator of the degree of eutrophication in the water, so accurately obtaining and predicting the concentration of chlorophyll-a can provide a theoretical basis for protecting the marine environment. In the experiment, remote sensing images obtained by the moderate resolution imaging spectrometer were used as data sources, and images of chlorophyll-a concentration in the same water area were taken as the true relative value. The convolution neural network was used to establish the relationship model between remote sensing reflectance and chlorophyll-a concentration, and then the inversion of marine chlorophyll-a concentration was realized. The experimental procedure starts with pre-processing the global ocean reflectance data (band combinations of 412, 469, 488, 547, 667 nm) and chlorophyll-a concentration data in 2020 by multiplicative amplification and logarithmic transformation. Then, a convolutional neural network inversion model for the concentration of marine chlorophyll-a was constructed based on the data set of the water area at the boundary of the Pacific Ocean and the Indian Ocean in January 2020, which was divided into the training set and validation set. The coefficient of determination (R2), Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) were used as evaluation indicators to optimise the inverse model. Finally, the data of chlorophyll-a concentration from January to December 2020 were used as the test set to verify the inversion accuracy of the model. The results show that the proposed inversion model has an accuracy of R2=0.930, RMSE=0.130, MAE=0.102 and demonstrate that the inversion results of chlorophyll-a concentrations given by the model are in high agreement with ground truth values. It can be applied in inversion studies of global marine chlorophyll-a concentration based on remote sensing imagery.