DETR-Based Approach for Ionogram Automatic Scaling

The ionosphere is critical for radio communication, navigation, and deep space exploration, with ionograms serving as key tools for real-time ionospheric measurement. Deep learning has recently been applied to enhance ionogram analysis, however, challenges persist, including suboptimal accuracy, poor dataset annotation, lack of fine-grained Es-layer classification, and limited robustness in complex scenarios. This study addresses these issues by integrating the DETR model with ionogram detection, utilizing a dataset of 4,311 ionograms from the Chinese Academy of Sciences. A comparative analysis of DETR, YOLOv3, and Mask2Former, RTMDet-tiny, Faster-RCNN reveals that DETR improves object detection accuracy, precision and recall.