Intelligent Identification and Key Parameter Extraction of Middle and Upper Atmospheric Disturbances Based on All-sky Airglow Imaging Observations of the Chinese Meridian Project

To address the demand for efficient processing of massive airglow images in the Meridian Project, this study developed a machine-learning-based method for automatic identification and parameter extraction of Atmospheric Gravity Waves (AGWs) and Medium-Scale Traveling Ionospheric Disturbances (MSTIDs). A Convolutional Neural Network (CNN) classification model was employed to filter clear-night-sky images, achieving accuracies of 99% (OH airglow) and 96.9% (OI airglow). Wave structures were localized using a Fast Region-Based CNN with an Intersection-over-Union (IoU) value exceeding 75%. For AGWs, parameters including wavelength, propagation direction, and horizontal phase velocity were extracted via 2D Fourier transform, while Canny edge detection and linear fitting were applied to MSTIDs. Analysis of the extracted parameter dataset revealed long-term trends of atmospheric waves: At the Dandong station (40.0°N, 124.0°E), OH airglow observations showed a bimodal seasonal distribution of AGW occurrence, with peaks during both winter and summer, with propagation directions being predominantly southwestward in winter and northeastward in summer. At the Xinglong station (40.2°N, 117.4°E), 94% of MSTID events detected via OI airglow exhibited southwestward propagation (azimuths of 200°～230°). These statistical characteristics align with established patterns in the literature, validating the reliability of the dataset. This tool resolves the inefficiency and subjectivity of traditional manual analysis, providing robust data support for long-term atmospheric wave studies. The associated algorithms and datasets will be open-sourced.