Sea surface temperature inversion model for infrared remote sensing images based on deep neural network.

• We propose a new sea surface temperature inversion model. • The model is based on deep neural network. • This model is more accurate than traditional method. • The model is applied to different sea areas and we get good results. The traditional sea surface temperature (SST) inversion model has a complicated parameter fitting process and poor adaptability in different sea areas. This paper presents an infrared remote sensing inversion model of SST based on deep neural network to refine the situation. The training data are the moderate-resolution imaging spectroradiometer (MODIS) infrared remote sensing data on sunny days and measured data from buoy in Bohai. The accuracy of inversion results is analyzed, the determination coefficient of inversion and measured values is 0.98, the standard error is 0.71 °C and the mean absolute deviation is 0.85 °C, the results show good accuracy of the model. The accuracy of Bohai SST inversion results is compared with SST products from the MODIS sensors and the inversion model is applied to other sea areas, demonstrating the credibility and portability of the model. The data experiments in this paper prove the feasibility of the model, which provides ideas for global SST inversion. [ABSTRACT FROM AUTHOR]