基于YOLO算法及风云静止卫星云图的台风识别和中心定位算法研究OA

Typhoon is a common disastrous weather system,accompanied by drastic weather changes,often causing disasters such as heavy rain,strong wind,tornadoes,and hail.The monitoring and research of typhoons are one of the key issues in meteorological disaster prevention and mitigation.Based on the best track dataset of tropical cyclones from the China Meteorological Administration(CMA)Tropical Cyclone Data Center in 2022 and the level-1 full disk nominal data of Fengyun-4A satellite,this study transformed the satellite raw data into satellite cloud images and constructed a sample annotation dataset for typhoons.Then,the YOLO v3 and YOLO v11(You Only Look Once version 3,11)object detection algorithms were used to study typhoon detection and center localization.Results show that YOLO v11 demonstrates outstanding performance in the typhoon detection and localizationtask,significantly improving the detection accuracy compared to previous studies.The precision and recall reached 98.71%and 99.34%respectively,which are significantly higher than 95.49%and 87.59%of YOLO v3.The average longitude deviation of typhoon center localization in this paper is 0.06°,the average latitude deviation is 0.07°,and the average distance deviation is 11.04 km.This paper further explores three key factors affecting the accuracy of typhoon center localization.The study found that the localization deviation of infrared cloud images is 25.1%lower than that of visible-light cloud images,the existence of typhoon eyes improves the localization accuracy by 17.5%compared to typhoons without eyes,and the number of coexisting typhoons in the cloud image has no significant impact on the location results.The path simulation experiments based on typhoon sequences show that YOLO v11 can maintain stable performance in typhoon events of different intensities and structures,effectively detecting and precisely locating typhoon sequences.