SAR-to-Optical image translation

Agenda
Reason behind image-to-image translation
- Problems with manufacturing
Structure of neural network
- Vanilla GAN and S-Cycle GAN
Experiments conducted
- Image translation
- Cloud removal
2022

Image-to-Image
Problems with optical images:
- Weather conditions
- Hours of darkness
2022
https://avatars.mds.yandex.net/get-zen_doc/176438/pub_5a8545c81410c32cdb85b939_5a8546c455876b90a66af9f2/scale_1200

Image-to-Image
Problems with synthetic aperture radar (SAR) images:
- Grainy noise
- Lack of colours
2022
https://2.bp.blogspot.com/-e1eA2Va5i8I/WDN8DbaNZHI/AAAAAAAACDE/ly7lKndXyUY5J6tCH0Wcpy9HW31kn2kygCLcB/s1600/intensity.JPG

Image-to-Image
The way to mitigate all problems at once
2022
SAR images
Generated images
Real optical images

GAN Structure
Discriminator (D) identify images
Generator (G) create image from noise
2022
https://qiita-user-contents.imgix.net/https%3A%2F%2Fqiita-image-store.s3.ap-northeast-1.amazonaws.com%2F0%2F567823%2F8c19ec3d-693d-e5c1-070c-ce69f1a16d58.jpeg?ixlib=rb-1.2.2&auto=format&gif-q=60&q=75&w=1400&fit=max&s=06c4a34575c9c4963f0f7076eb8f4b72

GAN Structure
Red rectangles: scenes for training
Blue rectangles: scenes for test
2022

GAN Structure
Division into smaller patches
Translation from SAR to optical
Stitching pieces the image
2022

Experiments and results
Image translation:
Example of ready picture
Example of translated patch
2022
SAR images
Generated images
Real optical images

Experiments and results
2022
Comparison of various approaches

Experiments and results
Cloud removal experiment
2022

Summary
Worthwhile and important problem has found it’s solution
Modern and elegant methods were used
Mind-blowing results were presented
2022