A Computer Vision Method For Precipitation Inversion With Satellite Cloud Images

Precipitation data is of great significance for agriculture,social activities and econ-omy.In terms of national meteorological disaster protection and regional climate research,there is an urgent need for precipitation data in a wide range of regions.In the current mete-orological operation,obtaining precipitation data mainly depends on the analysis of radar echo map by local weather radar station and the inversion of remote sensing algorithm of meteorological satellite sensor.However,the number of domestic land meteorological stations is sparse and uneven,and the long-term fixed radar detection in the vast ocean or desert is even more difficult.The existing open source remote sensing algorithms can't help to retrieve the precipitation data of different areas with low spatial and temporal reso-lution.In order to supplement the blind wide coverage and all-weather precipitation data,and provide a new perspective and method for high accuracy wide coverage precipitation data acquisition,this paper takes the satellite cloud image inversion precipitation method based on computer vision as the research topic,and mainly carries out the following work.1)This paper investigates the traditional algorithms and deep learning methods of computer vision in image segmentation,semantic segmentation and sequence prediction,discusses the related technologies of radar rainfall measurement and precipitation predic-tion,and combines the deep convolution network and long-term and short-term memory artificial neural network to solve the blind problem of wide coverage all day precipitation data.2)This paper analyzes the format and quality problems of many training samples ob-tained in this paper.According to the results of the problem analysis,the training samples such as satellite remote sensing cloud images are processed,and the high-quality label sample set with compound training requirements is made.3)Aiming at the nonlinear relationship between cloud features captured by meteo-rological satellite remote sensing and radar echo value,a satellite cloud image inversion model based on computer vision method is proposed.The cloud features are abstracted from the model,and the parameters in the model are trained to fit the nonlinear relation-ship between satellite cloud image pixels and radar echo value.The method proposed in this paper is superior to the traditional algorithms in coverage,portability,observation time interval,clarity and accuracy.4)In order to make up for the blind precipitation data better,this paper proposes a wide coverage precipitation evolution prediction composite model based on long-term and short-term memory artificial neural network.Compared with the traditional optical flow extrapolation prediction,it has better motion prediction accuracy,and can more accurately predict the precipitation movement trend and the evolution of precipitation value.Through the deep convolution neural network for satellite cloud image retrieval radar echo value accuracy,and the long-term and short-term memory artificial neural network for precipitation distribution change prediction accuracy,through the mean square error and balance f parameter for the accuracy of the two models,it shows that the composite model proposed in this paper can better complete the all-weather,wide coverage precipi-tation inversion task,and achieve the goal Blinding of precipitation data.