Remote Sensing Sea Ice Classification And Elevation Retracking Calibration Based On Machine Learning

As a very common and extremely important type of natural objects,sea ice plays a vital role in global climate change,polar resource development and maritime shipping.In order to understand the impact of sea ice on the world,the study of sea ice is closely concerned by relevant researchers at home and abroad.The analysis of sea ice related parameters based on remote sensing data has become a hot research topic.Therefore,our paper focuses on remote sensing data of sea ice and uses machine learning algorithms to conduct in-depth research on sea ice classification and sea ice elevation retracking calibration methods.In the direction of sea ice classification,most of the traditional methods use polarization and texture features,but do not use spatial concomitant relations of sea ice.In addition,existing sea ice classification methods based on spatial concomitant relations all need to formulate a rule set of sea ice spatial concomitant relations.However,the established rule set not only cannot completely include all concomitant relations,but also needs to manually set parameters based on experience.To solve these problems,we propose a deep model based on cascaded concomitant relation encoder.First,the model uses concomitant relation encoder to extract the spatial relations of sea ice as features and connects the extracted concomitant features with the bands of sea ice SAR remote sensing image as the input vectors.Then,the concomitant relation encoder randomly selects a very small number of input vectors as training samples for sea ice training.Finally,the deep model is continuously trained through the cascade framework to achieve the pixel-level sea ice classification.Experiments prove the effectiveness of the proposed deep model.In the direction of sea ice elevation retracking calibration,sea ice elevation data are mainly obtained from airborne radar and satellite radar.Although the sea ice elevation measured by airborne radar has great accuracy,it is extremely susceptible to environmental influences,especially in the Antarctica and Arctic.Each observation process needs to consume a lot of manpower and material resources.Moreover,bad weather makes the airborne radar bump up and down during the measurement and transportation,which is easy to damage the hard disk that stores the data.Satellite radar with strong survivability can obtain sea ice elevation data all-time and all-weather,regardless of environmental restrictions.However,due to the influence of natural phenomena and its own instruments,the measurement results of satellite radar altimeter contain errors.In order to improve the accuracy of sea ice elevation measured by satellite radar,we propose using machine learning algorithm to calibrate the sea ice elevation.Firstly,in order to use machine learning algorithm to achieve sea ice elevation retracking calibration,we need to produce relevant data sets.In this paper,five characteristic parameters are extracted from the echo received by the satellite radar altimeter as the characteristic values of the data set.The five characteristic parameters are the pre-pulse noise power,peak power,range gate number,slope and pulse width at half-power point respectively.Then,the sea ice elevation differences measured by airborne radar and satellite radar of the corresponding echoes are used as the target values of the data set.Finally,the data are input into the support vector regression model of LIBSVM,so as to obtain the differences of sea ice elevation and realize sea ice elevation retracking calibration.We conduct experiments on the remote sensing data of sea ice elevation in Greenland,and experiments prove the effectiveness of the machine learning algorithm in realizing sea ice elevation retracking calibration.In summary,using machine learning algorithms to study sea ice remote sensing data can effectively improve the accuracy of sea ice related parameters,so that people can more comprehensively understand global climate change and improve human production and life.