Research And Implementation Of Thunderstorm Gale Recognition Algorithm Based On Temporal And Spatial Data

Thunderstorm gale is a kind of severe weather disaster phenomenon in the field of severe convection weather.It belongs to small and medium-scale meteorological weather,which has the characteristics of rapid generation,high intensity,varied,strong damage and difficult prevention,so it has a relatively strong threat to people's social life and life safety.The identification and prediction of thunderstorm gale is a more challenging problem than the medium and long scale meteorological weather.Therefore,domestic and foreign experts have done a lot of research on the identification and prediction of thunderstorm gale.After the rise of deep learning,the use of deep learning technology for the identification and prediction of thunderstorm winds has gradually come into the field of vision.In this paper,a sample set of radar echo three-dimensional puzzle data from 2008 to 2018 in a province is used to study the task of thunderstorm and gale recognition.After analyzing the advantages and disadvantages of deep learning target detection algorithm,the YOLOv3 algorithm,which has a great advantage in detection speed and maintains a high precision,is selected as the basic research model of this paper.In this paper,the PAN method is combined with the original output layer structure of the model to construct a double pyramid structure of the output layer,and then optimized according to the characteristics of the range of thunderstorms and gales,and a better output layer structure is obtained.Radar echo threedimensional puzzle data is different from ordinary images.It has three-dimensional spatial characteristics.Therefore,the three-dimensional convolution method commonly used in video images or medical images is introduced to construct new C3 D basic components and join the backbone network through the backbone network and output layer The optimized structure is combined to propose the YOLO-S model.Then,after studying the dependence of radar echo timing,the spatiotemporal feature extraction structure is integrated into the YOLO-S model,and the fused model is called the ST-YOLO model.Finally,the improved method was compared with the original model,and experimental analysis verified that the optimization of the model was effective.Compared with the traditional method,the models proposed have better effects.In this paper,a thunderstorm gale recognition system based on target detection is built by using the designed target detection model.The system has two modes: batch processing and real time.It can quickly perform thunderstorm and gale recognition processing of historical samples through simple configuration,and it can also detect the latest time in real time.The three-dimensional puzzle data of the radar echo of the second time.The system can output the meteorological products of thunderstorm gale identification for the meteorological personnel,and provide effective reference basis.