10:45 AM - 12:15 PM
[MGI31-P05] Heavy Snow Cloud Detection in Satellite Images Based on Semi-Supervised Image Segmentation
Keywords:Machine Learning, Image Segmentation, Satelite Cloud Images, Semi-Supervised Learning, Supervised Learning
Long-term analysis of snowfall in Antarctica is important to figure out the surface mass balance of Antarctica. However, it is difficult to measure snowfall intensity over Antarctica for various reasons. Although, the polar orbiting satellites, such as NOAA, continuously image the clouds over Antarctica. Hence, It is crucial to estimate snowfall from satellite cloud images.
In this work, we used an image segmentation method based on a semi-supervised binary classification to detect heavy snow clouds from satellite images automatically. Also, we adopted supervised classification as a baseline method to verify that the method based on semi-supervised learning is useful. In this experiment, we utilized satellite images from AVHRR (The Advanced Very High-Resolution Radiometer) on the NOAA-18,19 satellites. We used CoSPA (Cost-effective Segmentation with Partial Annotations)[1], an image segmentation technology based on semi-supervised learning. The feature of CoSPA is that it uses partial annotations which cut off from the original satellite images. Besides, for the baseline experiment, we adopted U-Net[2] for the baseline experiment, an encoder-decoder model based on supervised learning. For the dataset, we applied the cloud images observed in 2008 by the AVHRR on the NOAA-18 and NOAA-19 satellites[3]. The Jacquard index (Intersection over Unions, IoU) and the Dice Coefficient is the evaluation method to evaluate the results from the above experiments.
[1] Keiichi Nakanishi, Ryoya Katafuchi, Terumasa Tokunaga, “ CoSPA: Cost-effective Image Segmentation from Partial Annotations based on deep PNU learning ”. in preparation
[2] Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. Springer International Publishing, 2015.
[3] Suzuki, Kazue, et al. "Identifying Snowfall Clouds at Syowa Station, Antarctica via a Convolutional Neural Network." Advances in Artificial Intelligence: Selected Papers from the Annual Conference of Japanese Society of Artificial Intelligence (JSAI 2020) 34. Springer International Publishing, 2021.
In this work, we used an image segmentation method based on a semi-supervised binary classification to detect heavy snow clouds from satellite images automatically. Also, we adopted supervised classification as a baseline method to verify that the method based on semi-supervised learning is useful. In this experiment, we utilized satellite images from AVHRR (The Advanced Very High-Resolution Radiometer) on the NOAA-18,19 satellites. We used CoSPA (Cost-effective Segmentation with Partial Annotations)[1], an image segmentation technology based on semi-supervised learning. The feature of CoSPA is that it uses partial annotations which cut off from the original satellite images. Besides, for the baseline experiment, we adopted U-Net[2] for the baseline experiment, an encoder-decoder model based on supervised learning. For the dataset, we applied the cloud images observed in 2008 by the AVHRR on the NOAA-18 and NOAA-19 satellites[3]. The Jacquard index (Intersection over Unions, IoU) and the Dice Coefficient is the evaluation method to evaluate the results from the above experiments.
[1] Keiichi Nakanishi, Ryoya Katafuchi, Terumasa Tokunaga, “ CoSPA: Cost-effective Image Segmentation from Partial Annotations based on deep PNU learning ”. in preparation
[2] Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. Springer International Publishing, 2015.
[3] Suzuki, Kazue, et al. "Identifying Snowfall Clouds at Syowa Station, Antarctica via a Convolutional Neural Network." Advances in Artificial Intelligence: Selected Papers from the Annual Conference of Japanese Society of Artificial Intelligence (JSAI 2020) 34. Springer International Publishing, 2021.