Identification of Electromagnetic Interference Echoes from Weather Radar Data Using Deep Learning-Based Object Detection

With the growing complexity of electromagnetic environments, electromagnetic interference (EMI) poses a significant threat to the quality of weather radar data. However, identifying EMI echoes presents substantial challenges due to their small target sizes and diverse morphological characteristics, frequently leading to misclassification and missed detections in practical application. To address these limitations, a deep learning-based EMI echo identification algorithm is proposed in this study. The algorithm enhances You Only Look Once (YOLO) object detection framework through several innovations: integrating Switchable Atrous Convolution (SAConv) into the C2f module of the backbone network to improve multi-scale feature representation, combining the Large Separable Kernel Attention (LSKA) mechanism with the Spatial Pyramid Pooling Fast (SPPF) module to refine feature extraction, incorporating the Convolutional Block Attention Module (CBAM) into the segmentation head to emphasize critical features while suppressing irrelevant interference, and employing dynamic Wise Intersection over Union (Wise-IoU) v2 as the bounding box regression loss to improve localization accuracy. With these architectural improvements, a novel YOLO Radar Electromagnetic Interference Echo (YOLO-REIE) model is developed and trained on a dataset of EMI echo samples from the Suizhou S-band weather radar. Comprehensive evaluation demonstrates that YOLO-REIE achieves superior performance, with a precision of 95.76%, a recall of 95.31%, and a mean Average Precision at IoU=0.50 (mAP50) of 93.32%, significantly outperforming other object detection models. The research findings provide an efficient technical approach for the identification of EMI echoes, thereby guaranteeing the quality and reliability of weather radar data.