Ground-based Cloud Image Segmentation Based On Deep Learning

Clouds,as an important meteorological element,have been widely studied and applied in the past few decades.However,due to the change of clouds in the sky over time and illumination,the shape characteristics of cloud will change.Therefore it is still a challenging task to segment cloud image accurately.At present,cloud image segmentation algorithms based on traditional methods are sensitive to the selection of parameters,lack certain robustness.When the foreground and background of cloud images are more complicated,researchers need to do a lot of feature engineering to obtain better results.Therefore,the classical image segmentation method based on shape prior is not suitable for cloud image segmentation.Based on the analysis of related studies by scholars,we finds that convolutional neural networks have good feature extraction capabilities and are capable of complex image semantics.The paper mainly studies the application of deep learning in cloud image segmentation.In order to improve the accuracy of ground-based cloud image segmentation,we first proposes a symmetrical and densely connected convolutional neural network cloud image segmentation method for ground-based cloud image segmentation research.The proposed new network structure first extracts the ground-based cloud image features through ordinary convolutional layers,then further processes the feature maps through continuous densely connected blocks and upsampling modules,and finally combines the shallow network and deep network feature maps in parallel to achieve precise segmentation of ground-based cloud images.Then,in order to better use the algorithm based on deep learning in practical applications,we adopt a lightweight backbone network and propose a new multi-level feature reuse fusion module to reduce the amount of parameters and calculations of the cloud image segmentation model.The experimental results show that the model proposed in this paper has a good performance on cloud image segmentation tasks.