Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi-supervised learning algorithm

Chu, Yunpei; Zhang, Huiying; Li, Xia; Henneberger, Jan

doi:https://doi.org/10.5194/amt-18-2781-2025

Articles | Volume 18, issue 12

https://doi.org/10.5194/amt-18-2781-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/amt-18-2781-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 18, issue 12

Research article

|

27 Jun 2025

Research article |

| 27 Jun 2025

Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi-supervised learning algorithm

Yunpei Chu, Huiying Zhang, Xia Li, and Jan Henneberger

Data sets

Data for publication "Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi-supervised learning algorithm" Yunpei Chu et al. https://doi.org/10.5281/zenodo.14696359

Model code and software

Models for publication "Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi-supervised learning algorithm" Yunpei Chu et al. https://doi.org/10.5281/zenodo.14793334

Code for publication "Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi-supervised learning algorithm" Yunpei Chu et al. https://doi.org/10.5281/zenodo.14795274

Short summary

Our study improves ice crystal shape classification, key for understanding weather and climate. By adding unsupervised pre-training before supervised classification, our algorithm reduces manual labeling effort while maintaining high accuracy. It outperforms fully supervised models across datasets of varying sizes and categories, showing strong generalization ability. This method improves ice crystal classification techniques, making it adaptable to different environmental datasets.