基于SE-UNet的天气雷达电磁干扰杂波识别技术OA

The data quality issues caused by electromagnetic interference clutter account for more than 90％of all data quality problems in weather radar observations.Whenever these false data being incorporated into the base data,will introduce significant errors in precipitation estimation,backscatting analysis,and extrapolation of meteorological data.To better identify these abnormal echo returns and implement control measures to ensure the quality of weather radar observation data,this paper proposes an improved semantic segmentation approach based on deep learning,incorporating the SEBolock module into the U-Net model architecture to establish a new model,the SE-UNet model,for identifying electromagnetic interference clutter in weather radar.Based on the analysis and validation,the Mean Pixel Accuracy(MPA)and Mean Intersection-over-Union(MIoU)of the SE-UNet model both reached 99.9％,surpassing the performance metrics of SegNet,LinkNet,and U-Net models,which also having the lowest loss rate(Los).This indicates that the proposed model demonstrates high reliability in distinguishing electromagnetic interference clutter from precipitation returns.