Evaluation of Deep Learning-Based Neural Network Methods for Cloud Detection and Segmentation

This paper presents a systematic approach for accurate short-time cloud coverage prediction based on machine learning (ML) approach. Based newly built omnidirectional ground-based sky camera system, local training and evaluation data sets were created. These used to train several state-of-the-art deep neural networks object detection segmentation. For this purpose, the camera-generated full hemispherical image every 30 min over two months in daylight conditions with fish-eye lens. From set, subset of images was selected according various criteria. Deep networks, two-stage R-CNN architecture, trained compared U-net segmentation implemented by CloudSegNet. All chosen then evaluated situation.