| 摘要: |
| 珊瑚礁是海洋生态系统的重要组成部分，对保护海洋生物多样性以及维持海洋生态平衡具有重大意义。我国南海珊瑚岛礁自然资源丰富，准确、高效地提取珊瑚岛礁信息对南海岛礁监测、管理、规划与保护具有现实意义。本研究基于我国海洋一号C卫星（HY-1C）遥感数据，对西沙永乐群岛珊瑚礁信息进行了研究与分析，提出基于HY-1C遥感数据的珊瑚礁地貌分类体系。采用全卷积神经网络UNet模型，依次通过下采样、上采样操作提取西沙永乐环礁地貌特征，实现原始影像的像素级语义分割。结果表明：基于HY-1C数据建立的地貌分类体系对活珊瑚覆盖及珊瑚生长发育条件具有指示作用，提出的基于U-Net模型的珊瑚岛礁地貌信息自动提取方法，能够为我国南海珊瑚岛礁生态系统的全自动、大范围监测和评价提供相应理论基础，在珊瑚礁生态管理与评价中发挥关键作用。精度验证结果表明：U-Net模型可以有效提取珊瑚礁地貌信息，采用的地貌信息提取方法具备时空泛化能力，泛化精度高于80%。 |
| 关键词: 海洋物理学 海洋一号C卫星（HY-1C） 海岸带成像仪 珊瑚礁地貌分类 深度学习 U-Net 西沙永乐环礁 |
| DOI：10.3969/J.ISSN.2095-4972.2022.04.010 |
|
| 基金项目:海洋领域融合应用示范项目资助项目（2020010004）；“龙计划”第五期国际合作研究项目资助项目（57971） |
|
| Study on remote sensing information extraction from coral reefs around Xisha Yongle Islands by deep learning method |
| MA Zhenni,SONG Yan,ZOU Yarong,ZHU Haitian,CUI Songxue |
| (China University of Geosciences, Wuhan, Wuhan 430074, China;National Satellite Ocean Application Service, Beijing 100081, China) |
| Abstract: |
| Coral reef is an important part of the marine ecosystem and is of great significance to the protection of marine biodiversity and the sustainable marine ecological balance. Coral reefs in the South China Sea are full of natural resources. Accurate and efficient extraction of coral reef information is of practical significance for monitoring, management, planning and protection of the reefs in the South China Sea. Based on the remote sensing data of China's HY-1C satellite, the coral reef information of Yongle Islands in Xisha was analyzed, and a coral reef geomorphology classification system was proposed accordingly. The U-Net model of full convolutional neural network was used to extract the geomorphic features of Xisha Yongle Atoll by down-sampling and up-sampling operations in turn to realize the pixel-level semantic segmentation of the original image. The result shows that the geomorphological classification system established based on HY-1C data is indicative of live coral cover and growth conditions. The proposed automatic extraction method of coral reef geomorphological information based on U-Net model can provide corresponding theoretical basis for the fully automatic and large-scale monitoring and coral reef ecosystem evaluation in the South China Sea can play a key role in the ecological management. The verification results show that the adopted geomorphological information extraction method has the ability of spatial and temporal generalization accuracy higher than 80%. |
| Key words: marine physics HY-1C coastal zone imager geomorphological classification of coral reef deep learning U-Net Xisha Yongle Atoll |
