Deep Learning for Atmospheric Cloud Image Segmentation

As a result of the work, the main segmentation methods were analyzed when applied to images of atmospheric clouds obtained by remote sensing methods. It is proposed the approach which is a further development of the deep learning model based on CNN class U-net. The quality of cloud image segmentation using various methods based on the Intersection over Union (IoU) criterion is presented. The evaluations of the advantages and disadvantages of the proposed segmentation method are provided. Experimental studies have shown the feasibility of using neural network segmentation with deep learning, which allows to localize the clouds in the image. The results can be used in the systems of monitoring and classification of the regions of Ukraine on the distribution of cloud masses in the seasons based on images of satellite weather maps.