Shallow Cloud Classification using Deep Learning and Image Segmentation

Pp: 153-174 (22)

* (Excluding Mailing and Handling)

Abstract

Shallow clouds play a significant role in the earth’s radiation balance, but they’re still poorly represented in climatic models. Our project analyzes the cloud images taken from satellites and attempts to build a deep learning model to classify cloud patterns. This will help us to identify the cloud formations and help improve the earth’s climate understanding. We will use various deep learning and image segmentation techniques like UNet to produce a model which can classify the shallow layers of clouds into various labels (fish, flower, gravel, and sugar). Various data augmentation techniques are implemented to improve the proposed model. Additionally, transfer learning is implemented by using ResNet backbones to improve the performance of the segmentation model. This will help gain insights into the matter of shallow cloud effects on the earth’s climate, there by helping in the development of next-gen climate models without having to go through the tedious task of classifying the clouds present in the images first.

Keywords: Deep Learning, Image Segmentation, RAdam, Shallow Clouds, UNet.

Cite as