Rapid monitoring of surface water based on remote sensing data and DeepLabv3+ model
-
摘要: 地表水体监测对于水资源保护具有重要的参考价值。该文以2013—2022年的国产高分一号(GF-1)系列遥感影像为数据源,发展了一种基于深度学习模型DeepLabv3+的像素级地表水体遥感提取方法。在北京市密云区的实验结果表明,该方法可快速获取多期次像元尺度的地表水时空分布,提取结果与真实空间分布基本一致; 与随机森林算法、支持向量机算法和最大似然法等常规分类算法提取结果进行对比,所提方法的精确率和召回率分别达99.22%和98.01%,水体提取精度较高。通过长时间序列监测,2013—2022年间密云区地表水体面积经过持续性减小→增加→保持稳定3个过程。该方法提取精度和效率满足区域级水体空间范围变化监测的需求,在区域地表水资源遥感快速监测和生态评价等领域具有广阔的业务应用前景。
Surface water monitoring can provide important references for water resource protection. Using 2013-2022 remote sensing images from the domestic high-resolution GF-1 constellation, this study developed a pixel-scale method for surface water information extraction based on the DeepLabv3+deep learning model. The experimental results of derived in Miyun District of Beijing indicate that the proposed method can quickly obtain multiple phases of pixel-scale spatiotemporal distributions of surface water, with the extraction results roughly consistent with actual spatial distribution. Compared to conventional classification algorithms such as random forest, support vector machine, and maximum likelihood, this method exhibited extraction precision and recall of 99.22% and 98.01%, respectively, demonstrating high accuracy in water information extraction. The long-term serial monitoring results indicate that the surface water area evolved from a continuous decrease to an increase and then to stabilization from 2013 to 2022. Since the extraction accuracy and efficiency can meet the demand for the monitoring of the spatial changes in regional water bodies, the proposed method enjoys broad prospects for practical application in the fields of remote sensing-based rapid monitoring and ecological assessment of regional surface water resources.
-
关键词:
- 地表水体 /
- 高分一号 /
- DeepLabv3+ /
- 快速遥感监测 /
- 精度评价
-
Key words:
- surface water /
- GF-1 satellite /
- DeepLabv3+ /
- rapid remote sensing monitoring /
- accuracy evaluation
-
-
计量
- 文章访问数: 44
- PDF下载数: 10
- 施引文献: 0
下载: