Instance Segmentation of Auroral Images for Automatic Computation of Arc Width

By Chuang Niu, Qiuju Yang, Shenghan Ren, Haihong Hu, Desheng Han, Zejun Hu, and Jimin Liang.

Introduction

A fully automatic method for computing auroral arc width based on Marsk R-CNN is implemented in this project, and the related paper is submitted to GRSL. More details will be described.

Installation (Tested on Ubuntu 16.04)

This project is based on Mask R-CNN, PyTorch 1.0, and Python 3.5.

1. Install the Mask R-CNN benchmark.
2. Replace the maskrcnn-benchmark with the aurora-maskrcnn in this project.

3. git clone https://github.com/niuchuangnn/aurora-maskrcnn.git
   cd ~/aurora-maskrcnn
   python3 setup.py build develop

Demo

Run demo:

   cd ~/aurora-maskrcnn
   python3 ./Aurora/demo.py

It will output the following results:

1. Original image:

2. Detection results of one-stage inference process:

3. Detection results of rotated image:

4. Detection results of two-stage inference process:

5. Predicted normal of aurora arcs:

6. Intensity vs. zenith-angle curve:

License

maskrcnn-benchmark is released under the MIT license. See LICENSE for additional details.