SegCloud: a novel cloud image segmentation model using a deep convolutional neural network for ground-based all-sky-view camera observation

Abstract

Abstract. Cloud detection and cloud properties have substantial applications in weather forecast, signal attenuation analysis, and other cloud-related fields. Cloud image segmentation is the fundamental and important step in deriving cloud cover. However, traditional segmentation methods rely on low-level visual features of clouds and often fail to achieve satisfactory performance. Deep convolutional neural networks (CNNs) can extract high-level feature information of objects and have achieved remarkable success in many image segmentation fields. On this basis, a novel deep CNN model named SegCloud is proposed and applied for accurate cloud segmentation based on ground-based observation. Architecturally, SegCloud possesses a symmetric encoder–decoder structure. The encoder network combines low-level cloud features to form high-level, low-resolution cloud feature maps, whereas the decoder network restores the obtained high-level cloud feature maps to the same resolution of input images. The Softmax classifier finally achieves pixel-wise classification and outputs segmentation results. SegCloud has powerful cloud discrimination capability and can automatically segment whole-sky images obtained by a ground-based all-sky-view camera. The performance of SegCloud is validated by extensive experiments, which show that SegCloud is effective and accurate for ground-based cloud segmentation and achieves better results than traditional methods do. The accuracy and practicability of SegCloud are further proven by applying it to cloud cover estimation.