| 引用本文： |
-
文莉莉,赖俊翔,邬满,师德强.广西红树林遥感智能提取及其碳储量估算方法研究[J].广西科学,2026,33(2):243-257. [点击复制]
- WEN Lili,LAI Junxiang,WU Man,SHI Deqiang.Remote Sensing Intelligent Extraction of Mangroves in Guangxi and Carbon Storage Estimation Methods[J].Guangxi Sciences,2026,33(2):243-257. [点击复制]
|
|
| 本文已被：浏览 127次 下载 25次 |
码上扫一扫！ |
| 广西红树林遥感智能提取及其碳储量估算方法研究 |
|
文莉莉1, 赖俊翔1, 邬满1,2, 师德强1
|
|
|
| (1.广西科学院, 广西海洋科学院(广西红树林研究中心), 广西近海海洋环境科学重点实验室, 广西人机交互与智能决策重点实验室, 广西北部湾碳汇与低碳工程研究中心, 广西南宁 530007;2.广西科学院, 广西海洋实验室, 广西南宁 530004) |
|
| 摘要: |
| 针对红树林生态系统空间分布复杂、传统遥感提取方法精度受限及区域碳储量本底不清等问题，本文提出一种融合动态状态空间模型与局部注意力机制(LAM)的红树林智能提取方法，并实现广西海岸带红树林生态系统碳储量估算。该方法以U-Net为基础架构，针对红树林斑块长距离连通与边缘易混淆的特性，引入动态状态空间模块(DSSM)建模全局上下文依赖，结合局部注意力机制增强边缘细节感知，并嵌入多尺度动态选择机制(SKNet)提升模型对多尺度目标的鲁棒性。在此基础上，耦合高分辨率遥感提取结果与地面样方调查数据，构建包含植被生物量、沉积物及凋落物的红树林生态系统碳储量估算模型。实验结果表明，本文提出的改进模型在红树林提取任务中表现优异，总体精度(OA)达到96.3%，Kappa系数为0.90，显著优于基础U-Net和DeepLab V3+模型，可有效解决复杂海岸带背景下的精细分割难题。另外，本文基于智能提取结果与野外调查数据，得到广西钦州、北海、防城港3地的红树林生态系统碳储量分别为8.823×105、1.155×106、5.875×105 Mg C。本文可为区域蓝碳资源的科学管理、生态保护及碳交易市场建设提供坚实的数据支撑与技术范式。 |
| 关键词: 红树林 深度学习 遥感提取 动态状态空间模型 局部注意力机制 动态选择机制 碳储量估算 |
| DOI：10.13656/j.cnki.gxkx.20260603.004 |
| 投稿时间：2026-02-03修订日期：2026-02-20 |
| 基金项目:广西重点研发计划项目(桂科FN2600640417)资助。 |
|
| Remote Sensing Intelligent Extraction of Mangroves in Guangxi and Carbon Storage Estimation Methods |
|
WEN Lili1, LAI Junxiang1, WU Man1,2, SHI Deqiang1
|
| (1.Research Center for Carbon Sink and Low-Carbon Engineering in the Beibu Gulf of Guangxi, Guangxi Key Laboratory of Human-machine Interaction and Intelligent Decision, Guangxi Key Laboratory of Marine Environmental Science, Guangxi Academy of Marine Sciences (Guangxi Mangrove Research Center), Guangxi Academy of Sciences, Nanning, Guangxi, 530007, China;2.Guangxi Laboratory of Oceanography, Guangxi Academy of Sciences, Nanning, Guangxi, 530004, China) |
| Abstract: |
| In response to the complex spatial distribution of mangrove ecosystems,the limited accuracy of conventional remote sensing extraction methods,and the unclear background of regional carbon storage,this paper proposes an intelligent extraction method for mangroves that integrates dynamic state space models and Local Attention Mechanisms (LAM),and estimates the carbon storage of mangroves in the coastal zone of Guangxi.This method is based on the U-Net architecture.Considering the long-distance connectivity of mangrove patches and the confusion of their edges,this study introduces a Dynamic State Space Module (DSSM) to model global context dependencies and a local attention mechanism to enhance edge detail perception,and embeds a multi-scale dynamic selection mechanism (SKNet) to improve the model robustness to multi-scale targets.On this basis,this study couples high-resolution remote sensing extraction results with ground plot survey data to construct a mangrove carbon storage estimation model involving vegetation biomass,sediment,and litter.Experimental results show that the improved model proposed in this study performs excellently in the task of mangrove extraction,with the Overall Accuracy (OA) of 96.3% and a Kappa coefficient of 0.90,significantly outperforming the basic U-Net and DeepLab V3+ models,effectively solving the problem of fine segmentation in complex coastal backgrounds.Additionally,using the intelligent extraction results and field investigation data,this paper systematically estimates the carbon storage of mangrove ecosystems in Qinzhou,Beihai,and Fangchenggang,Guangxi,which is 8.823×105,1.155×106,and 5.875×105 Mg C,respectively.This pater provides solid data support and technical paradigms for the scientific management,ecological protection,and carbon trading market construction of regional blue carbon resources. |
| Key words: mangrove deep learning remote sensing extraction dynamic state space model local attention mechanism dynamic selection mechanism carbon storage estimation |
