A Self-supervised Approach for Cloud Image Analysis

Clouds play a significant role in the global climate system, influencing the energy balance of the Earth’s atmosphere through absorption, reflection, and scattering of incoming shortwave solar radiation, and absorption and emission of outgoing longwave radiation. Different types of clouds result from different atmospheric processes, and accurately identifying and analyzing cloud coverage is essential for understanding the Earth’s radiative budget. While traditional computer vision methods and newer deep neural network (DNN) approaches have been used for cloud segmentation and categorization, these methods often rely on subjective human labeling of cloud or clear sky pixels for training. This can bias the algorithm’s performance or hinder the extraction of relevant information from sky images. In addition, labeling cloud images can be a tedious process. In this work, we present a self-supervised learning approach for autonomously extracting relevant features from sky images captured by ground-based cameras for the classification and segmentation of clouds. Our approach aims to overcome the limitations of traditional methods and DNNs by using self-knowledge distillation and regularization to learn features from the data without the need for human labeling. We evaluate the performance of our approach on two datasets and demonstrate its ability to accurately classify and segment clouds.

Recommended citation: Bhupendra A Raut, et al. (2023). "A Self-supervised Approach for Cloud Image Analysis." 103rd AMS Annual Meeting. https://ams.confex.com/ams/103ANNUAL/mediafile/Manuscript/Paper413343/AMS2023_Abstract-413343.pdf