《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (4): 1310-1316.DOI: 10.11772/j.issn.1001-9081.2023040453

• 多媒体计算与计算机仿真 • 上一篇    

基于改进Res-UNet的昼夜地基云图分割网络

王铂越, 李英祥(), 钟剑丹   

  1. 成都信息工程大学 通信工程学院,成都 610200
  • 收稿日期:2023-04-21 修回日期:2023-07-05 接受日期:2023-07-05 发布日期:2023-12-04 出版日期:2024-04-10
  • 通讯作者: 李英祥
  • 作者简介:王铂越(1999—),男,河南洛阳人,硕士研究生,主要研究方向:智能图像处理、人工智能
    钟剑丹(1985—),男,甘肃张掖人,讲师,博士,主要研究方向:智能图像处理、人工智能。
  • 基金资助:
    四川省科技计划项目(2023YFS0428)

Segmentation network for day and night ground-based cloud images based on improved Res-UNet

Boyue WANG, Yingxiang LI(), Jiandan ZHONG   

  1. College of Communication Engineering,Chengdu University of Information Technology,Chengdu Sichuan 610200,China
  • Received:2023-04-21 Revised:2023-07-05 Accepted:2023-07-05 Online:2023-12-04 Published:2024-04-10
  • Contact: Yingxiang LI
  • About author:WANG Boyue, born in 1999, M. S. candidate. His research interests include intelligent image processing, artificial intelligence.
    ZHONG Jiandan, born in 1985, Ph. D., lecturer. His research interests include intelligent image processing, artificial intelligence.
  • Supported by:
    Science and Technology Plan of Sichuan Province(2023YFS0428)

摘要:

针对昼夜地基云图在分割中细节信息丢失、分割精度低等问题,提出一种基于改进Res-UNet(Residual network-UNetwork)的昼夜地基云图分割网络CloudRes-UNet(Cloud ResNet-UNetwork),整体采用编码器-解码器的网络结构。首先,编码器使用ResNet50提取特征,增强特征提取能力;其次,设计多级特征提取(Multi-Stage)模块,该模块结合分组卷积、膨胀卷积和通道打乱这3种技巧,获取高强度语义信息;再次,加入高效通道注意力(ECA?Net)模块,在通道维度上聚焦重要信息,加强对地基云图中云区域的关注,提高分割精度;最后,解码器使用双线性插值对特征进行上采样,提高分割图像的清晰度并减少目标和位置信息丢失。实验结果表明,与当前基于深度学习表现较好的地基云图分割网络(Cloud-UNet)相比,CloudRes-UNet在昼夜地基云图分割数据集上的分割准确率提升了1.5个百分点,平均交并比(MIoU)上升了1.4个百分点,更准确地获取了云量信息,对天气预报、气候研究和光伏发电等方面具有积极意义。

关键词: 地基云图, 语义分割, 深度学习, 高效通道注意力网络, ResNet50, Res-UNet

Abstract:

Aiming at the problems of detail information loss and low segmentation accuracy in the segmentation of day and night ground-based cloud images, a segmentation network called CloudResNet-UNetwork (CloudRes-UNet) for day and night ground-based cloud images based on improved Res-UNet (Residual network-UNetwork) was proposed, in which the overall network structure of encoder-decoder was adopted. Firstly, ResNet50 was used by the encoder to extract features to enhance the feature extraction ability. Then, a Multi-Stage feature extraction (Multi-Stage) module was designed, which combined three techniques of group convolution, dilated convolution and channel shuffle to obtain high-intensity semantic information. Secondly, Efficient Channel Attention Network (ECA?Net) module was added to focus on the important information in the channel dimension, strengthen the attention to the cloud region in the ground-based cloud image, and improve the segmentation accuracy. Finally, bilinear interpolation was used by the decoder to upsample the features, which improved the clarity of the segmented image and reduced the loss of object and position information. The experimental results show that, compared with the state-of-the-art ground-based cloud image segmentation network Cloud-UNetwork (Cloud-UNet) based on deep learning, the segmentation accuracy of CloudRes-UNet on the day and night ground-based cloud image segmentation dataset is increased by 1.5 percentage points, and the Mean Intersection over Union (MIoU) is increased by 1.4 percentage points, which indicates that CloudRes-UNet obtains cloud information more accurately. It has positive significance for weather forecast, climate research, photovoltaic power generation and so on.

Key words: ground-based cloud image, semantic segmentation, deep learning, Efficient Channel Attention Network (ECA-Net), ResNet50, Res-UNet (Residual network-UNetwork)

中图分类号: