Few shot skylight detection with StyleGAN2-ADA and YOLOv9-PSA

Lihong Li; Lingli Mu; YuQing He; Jiaxi Li; Lina Xian; Wei Zhang

doi:10.1117/12.3057559

28 January 2025 Few shot skylight detection with StyleGAN2-ADA and YOLOv9-PSA

Lihong Li, Lingli Mu, YuQing He, Jiaxi Li, Lina Xian, Wei Zhang

Author Affiliations +

Proceedings Volume 13506, Sixth International Conference on Geoscience and Remote Sensing Mapping (GRSM 2024); 1350625 (2025) https://doi.org/10.1117/12.3057559
Event: Sixth International Conference on Geoscience and Remote Sensing Mapping (ICGRSM 2024), 2024, Qingdao, China

Abstract

Skylights, as a unique and scarce type of pits on Mars, are significant for scientific research and engineering applications. Compared with visual interpretation to detect skylights, deep learning methods are more automatic and objective. However, existing deep-learning methods face low accuracy due to very few skylight samples and the model's inability to capture the subtle features that differentiate skylights from typical pits. To solve the problem, skylight samples generated with StyleGAN2-ADA network are used to expand the original dataset and not-skylight pits are used as negative samples for adversarial training. These improvements reduce the number of misidentified skylights to approximately 1.3% of the original. Furthermore, to enhance the feature extraction capability, we propose the YOLOv9-PSA model, leading to a 14.1% increase in precision. The results indicate that expanded dataset and the YOLOv9-PSA model significantly improve detection accuracy, with final metrics achieving a precision of 94.1%, a recall rate of 88.9%, and an F1 score of 91.4%. The method proposed in this paper provides a foundation for skylight detection of other celestial bodies.

(2025) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Lihong Li, Lingli Mu, YuQing He, Jiaxi Li, Lina Xian, and Wei Zhang "Few shot skylight detection with StyleGAN2-ADA and YOLOv9-PSA", Proc. SPIE 13506, Sixth International Conference on Geoscience and Remote Sensing Mapping (GRSM 2024), 1350625 (28 January 2025); https://doi.org/10.1117/12.3057559

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

;

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE