Research On Typhoon Recognition And Centering Based On Remote Sensing Data

With the development of economic globalization,it is bound to exacerbate the development of urbanization in our country.As a result,the urban population density in our country has rapidly increased.However,this has resulted in greater loss of life and safety of our citizens in the event of major natural disasters such as typhoons and torrential rains and at the same time huge social losses of wealth also become a problem.Coupled with the current deterioration of the ecological environment,human beings are more and more weak in front of natural disasters.Therefore,accurately forecasting natural disasters has become an urgent major issue at present.Among them,the forecast of typhoon has always been the focus of the field of meteorological forecast.Because typhoon is a highly disastrous weather system,and the typhoon must be forecasted to obtain the central position of the typhoon.Today,the monitoring of typhoons is mainly dependent on satellite remote sensing data,however,at present,the automatic positioning of the typhoon center in the world is still at a stage of exploration and research.In particular,there is not a mature positioning system for the automatic positioning of the typhoon-free typhoon.Different typhoon patterns require different methods to locate the typhoon center.Positioning process is complex and the results are not satisfactory.In order to solve these shortcomings,this research proposes a method of using deep learning network instead of traditional methods to identify and locate the center,and conducts experimental research.In this dissertation,according to the problem of lack of recognizing typhoon cloud system based on the traditional method,a deep learning method is proposed to replace the traditional method to recognize typhoon cloud system.The typhoon and the characteristics of FY-2 meteorological satellite data are deeply investigated,and the method of identifying the typhoon cloud system is analyzed.Based on this,an overall algorithm framework for the recognition of typhoon cloud system by deep learning is designed,including data preprocessing,model Training and parameter selection and testing and result analysis.According to the overall algorithm framework,an experiment was conducted to identify the typhoon cloud system by using the deep learning network.Satisfactory results were obtained on the recognition rates of the typhoon cloud system,and the feasibility of the scheme was verified.In this dissertation,according to the problem of traditional method of locating typhoon center complex and unable to locate some kind of typhoon,this paper proposes a deep learning location method based on historical path.The method of typhoon center location is deeply investigated.The traditional method of typhoon center location is analyzed.The simulation experiment of typhoon center is carried out.The satellite data are transformed and the experimental datasets are generated.Based on this,a new typhoon center positioning method is designed.According to the method,the historical path is combined with the Ins Faster R-CNN network to simulate the typhoon center positioning.The experimental results show that in all typhoons,the center has achieved satisfactory results on positioning,and finally verified the feasibility of this method.