Segmentation and Edge Detection for Ionogram Automatic Scaling

by Yijie Zheng, Xiaoqing Wang, Yefei Luo, Hao Tian, Ziwei Chen.

Code for paper Segmentation and Edge Detection for Ionogram Automatic Scaling.

Prerequisites

A new branch based on OpenMMLab 2.x and mmseg 1.x is now available!

File Structure

./
├── data
│   ├── BuildDataset.ipynb
│   └── IonoSeg
│       ├── img
│       ├── label
│       ├── mask
│       ├── rgbimg
│       ├── rgbmask
│       ├── splits0
│       │   ├── splits.ipynb
│       │   ├── test.txt
│       │   ├── train.txt
│       │   ├── train_val.txt
│       │   └── val.txt
│       └── viz
│           └── 20130401040700.png
├── finetune_MMSegv0.ipynb
├── Inference.ipynb
├── README.md
├── tools
│   └── test.py
└── work_dirs
    └── se4ionogram
        ├── pspnet_r50_ionogram_mmseg0.py
        └── pspnet_r50_ionogram_iou_3922_acc_9153.pth

Dataset

The Dataset we use is available on google drive: Iono4311.rar.

For more information of dataset processing, please visit BuildDataset.ipynb.

Config

The configuration of PSPNet is saved here.

Finetune

Finetune the model by running finetune_MMSegv0.ipynb.

Test

python tools/test.py ./work_dirs/se4ionogram/pspnet_r50_ionogram_mmseg0.py \
/home/ubuntu/mmsegmentation/work_dirs/se4ionogram/pspnet_r50_ionogram_iou_3922_acc_9153.pth \
--eval mIoU

Inference

Get the ionospheric parameters by running the notebook Inference.ipynb.

Models and results

Method	Background Weight	Download	mTPR	DH	DF	dfoF2 $\le$ 0.2MHz	dhF2 $\le$ 10km
PSPNet	0.10	model	0.8713	4.38	0.12	98.6	97.0
PSPNet	0.15	model	0.8814	4.63	0.112	98.3	97.0
PSPNet	0.20	model	0.8070	4.69	0.100	98.5	97.8
PSPNet+Canny	0.08	model	0.9415	3.05	0.100	97.7	98.6
PSPNet+Canny	0.02	model	0.9256	2.88	0.091	98.4	98.8
PSPNet+Canny	0.05	model	0.9021	2.82	0.084	99.1	98.7
PSPNet+Canny	0.10	model	0.8713	3.01	0.08	99.0	98.5
PSPNet+Canny	0.15	model	0.8814	4.63	0.096	97.9	98.3
PSPNet+Canny	0.20	model	0.8070	4.05	0.093	98.3	97.1

Citation

If you find our work useful for your research, please consider citing:

@INPROCEEDINGS{9955166,
  author={Zheng, Yijie and Wang, Xiaoqing and Luo, Yefei and Tian, Hao and Chen, Ziwei},
  booktitle={2022 International Conference on Machine Learning, Cloud Computing and Intelligent Mining (MLCCIM)},
  title={Segmentation and Edge Detection for Ionogram Automatic Scaling},
  year={2022},
  pages={115-120},
  doi={10.1109/MLCCIM55934.2022.00026}
}

Contact

Should you have any questions, please send email to 19211416@bjtu.edu.cn.

VoyagerXvoyagerx / Ionogram