期刊文献+

结合水域变化的土壤湿度监测

Soil Moisture Monitoring Combined with Water Area Change
下载PDF
导出
摘要 针对现有土壤湿度监测方法受地表粗糙程度与植被覆盖影响大,导致土壤水分反演精度有待提升的问题,以康平地区为研究区域,提出了一种基于改进U-Net水域自适应提取模型的土壤湿度反演方法。该方法首先根据辽宁省多年份Sentinel-2影像制作遥感影像水体智能提取数据集,并在预设不同网络深度的自适应U-Net模型上进行训练;然后基于颜色熵阈值对影像进行复杂程度判断;最后将研究区多时相水域面积与同期实测土壤湿度数据进行回归分析,构建研究区水域面积与土壤湿度之间的内在联系。选用测试集影像,分别从提取精度和提取速度两个方面验证该方法和U-Net模型的水域提取性能。该方法水域轮廓提取更加精细,训练及预测时间分别缩短了25.65%和32.19%。此外,文章以14个时相的康平地区水域面积数据及同期实测土壤湿度数据,基于三次多项式拟合构建土壤湿度反演方法。结果表明,该反演方法在10 cm深度的土壤湿度反演实验中R 2为0.7234;在40 cm深度的土壤湿度反演实验中R^(2)为0.5689。该方法能够较准确地反演区域土壤湿度,进而为农业大范围土壤湿度的监测提供支持。 Aiming at the problem that the existing soil moisture monitoring methods are greatly affected by surface roughness and vegetation cover,the inversion accuracy of soil moisture needs to be improved.Taking Kangping area as the research area,this paper proposes a soil moisture inversion method based on the improved U-Net water adaptive extraction model.Firstly,the Sentinel-2 images of Liaoning province are used to produce intelligent water extraction data sets of remote sensing images,and then train it on the adaptive U-Net model with different network depths.Then the complexity of the image is judged based on the color entropy threshold.Finally,the regression analysis is conducted between the multi temporal water area of the study area and the measured soil moisture data during the same period to establish the internal relationship between the water area of the study area and soil moisture.In this paper,test set images are selected to verify the performance of the proposed method and U-Net model in terms of extraction accuracy and extraction speed.It can be seen that the water contour extraction method is more precise,and the training and prediction time are shortened by 25.65%and 32.19%,respectively.In addition,a soil moisture inversion method is established based on cubic polynomial fitting using 14 time-phase water area data and measured soil moisture data in Kangping area during the same period.The results show that the R^(2) of the above inversion method is 0.7234 in the soil moisture inversion experiment at a depth of 10 cm.The R 2 of soil moisture inversion experiment at 40 cm depth is 0.5689.The method presented in this paper can accurately invert the regional soil moisture and provide support for the monitoring of large-scale agricultural soil moisture.
作者 朱洪波 张兵 刘佳典 宋伟东 李佳 ZHU Hongbo;ZHANG Bing;LIU Jiadian;SONG Weidong;LI Jia(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China;School of Geospatial Information Service,Liaoning Technical University,Fuxin,Liaoning 123000,China;Dalian Huangbohai Marine Surveying Data Information Co.Ltd.,Dalian,Liaoning 116000,China)
出处 《遥感信息》 CSCD 北大核心 2023年第5期23-30,共8页 Remote Sensing Information
基金 国家自然科学基金面上项目(42071343)。
关键词 多元回归模型 水域提取 深度学习 土壤湿度反演 U-Net multiple regression model water area extraction deep learning soil moisture inversion U-Net
  • 相关文献

参考文献11

二级参考文献114

共引文献126

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部