Research On Meteorological Radar Data Processing Based On Deep Learning Technology

Meteorological radar originated in the middle of the last century.After decades of development,it has become one of the most important tools in weather forecasting and disaster prevention.Up to now,the latest weather radar is Doppler pulse radar.It transmits and receives electromagnetic waves and accordingly generates weather radar image data based on the intensity of the reflected waves.In the past,the processing of meteorological radar image data with traditional methods has certain shortcomings in efficiency and accuracy.With the development of artificial intelligence technology in recent years,the method of processing image data using deep learning technology has been more widely used.Deep learning technology uses a specific designed artificial neural network to learn a large number of image data and grasp its characteristics,so that the image data can be effectively and accurately analyzed and processed.Compared with traditional methods,the application of deep learning technology can more easily and efficiently implement the processing and analysis of weather radar image data.Based on Convolutional Neural Network(CNN)and Gated Recurrent Unit(GRU),we designed and implemented a neural network for the processing and analysis of weather radar image data.This neural network model can learn the regularity of weather radar image data and predict its future changes.The main work and contributions are as follows:1.Based on the convolutional neural network,a pre-processing neural network is designed and implemented to extract the characteristics of weather radar image data.The neural network uses two convolution layers to perform convolution operations on the original meteorological radar image,and uses different convolution kernels to extract a variety of original image data features to form a feature image.The preprocessed feature images are used for further neural network training.Through this preprocessing neural network,the features of the original data are extracted and a feature image is formed to greatly reduces the amount of redundant data and make subsequent data processing faster and more accurate.2.Based on the Gated Recurrent Unit(GRU),a neural network with a three-layer GRU structure is designed to implement feature learning on the meteorological radar feature images arranged in time sequence after preprocessing.Via the three-layer GRU structure with supplementary network structures,such as a fully connected layer and a pooling layer,the neural network can accurately and efficiently learn the information of the feature image.By learning the information from feature images,it is then applied to analyzing,processing and predicting the weather radar image data.3.The above neural network is implemented on a computer with ordinary performance.The neural network is trained by using a large number of timing sequenced feature images obtained by preprocessing,so that the neural network learns the change rule of the feature image,and then predicts the future changes of weather radar images.Compared with related research,the neural network designed and implemented in this study has excellent performance with faster training progress.Even without a high-performance GPU,network training and image prediction can be quickly implemented using general GPU devices.Under the premise of ensuring performance,the requirements for hardware equipment is lowered,so that it can be applied to more generally.