Remote Sensing Inversion Models Of Sea Surface Temperature Based On Deep Learning Method

Sea surface temperature is an important Marine physical parameter,which has great significance in oceanography and meteorology research.The development of remote sensing technology provides a wide range of data acquisition approaches for large scale sea surface temperature.Deep learning has made remarkable achievements in the field of artificial intelligence such as speech recognition and natural language processing due to its excellent feature learning ability.Based on remote sensing technology and deep learning technology,this paper studies the establishment method of sea surface temperature inversion model,realizes the inversion of sea surface temperature,analyses and evaluates the accuracy of the inversion results and the portability of the model.This paper selected the MODIS remote sensing image data and the buoy measured data of sea surface temperature from 2013 to 2016 in the bohai sea area established inversion model and carries on the sea surface temperature inversion.The radiometric calibration and geometric correction of remote sensing data is carried out in order to obtain remote sensing parameter data such as band brightness temperature and atmospheric transmittance,etc.And the data is normalized.The data of remote sensing data and the measured data are composed of data sets of 776 groups,and the data is randomly arranged.According to the ratio of 6:2:2,the data was divided into 466 sets of training data,155 sets of verification data and 155 sets of test data,respectively used for model training,model parameter adjustment,and model precision verification.By correlation analysis,5 remote sensing parameters T31,T31-T32,sec?,?31·T31 and ?32·T32 are determined as model input parameters.Then,the inversion model based on deep neural network is established.Because the diffusion of gradients and the over-fitting which often exist in the training of deep neural network,this paper selected to join optimization strategies such as the activation function Relu,the Dropout algorithm and the improved optimization algorithm Adam to overcome.In order to determine the model parameters,such as the number of hidden layers and iteration times,the parameters of the model are set by default,and the experimental comparison is performed to complete the model parameter setting.According to the accuracy of the verification data adjusts the other parameters,the parameters of the inversion model are determined eventually.The trained model is used for inversion of sea surface temperature in bohai sea.Area.The precision analysis of the obtained inversion results shows that the inversion model based on deep learning in this paper has obtained high precision,and the error of inversion results in most cases is less than 1?.The feasibility and superiority of the model based on deep learning in the inversion of sea surface temperature are proved.by comparing the inversion model established in this paper with the inversion results of the traditional regression statistical model.At the last part of this paper,the inversion model based on deep learning is transplanted into the south China sea and the results show that the accuracy is still high,which validated the portability of the model.Therefore,the research results of this paper have important theoretical and practical value for remote sensing inversion of sea surface temperature.