Dataset for: Automatic Characterization of Multiple Mid-latitude Ionospheric Plasma Structures from All-sky Airglow Images using Deep Learning Technique

This repository contains the data used in the study "Automatic Characterization of Multiple Mid-latitude Ionospheric Plasma Structures from All-sky Airglow Images using Deep Learning Technique". The images were recorded by the multi-wavelength all-sky airglow imager at Hanle, India (32.77°N, 78.97°E; Mlat. ~24.1°N) at the O(¹D) 630.0 nm wavelength and span the period 2018-2025. Contents images/ train/ validation/ test/ labels/ train/ validation/ test/ calculated_parameters.xlsx images/ - preprocessed unwarped images in single-channel grayscale, cropped to 500×500 pixels. The images have undergone geospatial calibration, noise removal, and geometric unwarping; details of these steps are given in Mondal et al. (2019), https://doi.org/10.1016/j.asr.2019.05.047. labels/ -polygon boundary annotations of the plasma structures, one text file per image. The annotations were generated using the VGG Image Annotator (VIA) and exported in YOLO segmentation format. A single class is used, representing a plasma structure. calculated_parameters.xlsx -the propagation parameters derived for the test dataset, containing for each tracked structure the date, time (UT), Track ID, velocity estimates from the Minima, MNCC, and Optical Flow methods, the final filtered velocity, the quality flag, the tilt angle, and the propagation direction. Splits The dataset is partitioned date-wise rather than randomly, so that images from the same night do not appear in more than one split. The training set contains 1253 images and the validation set 251 images. The test set was held out entirely from model development. Naming convention Files are named as WS_ddmmyyyy_hhmmss.extension, where the date is followed by the time in Universal Time (UT). Each label file shares the stem of its corresponding image, so images/train/WS_ddmmyyyy_hhmmss.jpeg pairs with labels/train/WS_ddmmyyyy_hhmmss.txt.