摘要
通过神经网络和机器学习的方法建立遥感影像的光谱信息与土壤湿度之间的模型,采用遥感手段大范围预测地表土壤湿度。以"天宫二号"2016年9月24日宽波段成像仪采集的可见光近红外谱段影像作为模型输入,选取与"天宫二号"影像相同采集时间和经纬度的SMAP/Sentinel-1 L2土壤湿度产品作为输出,分别通过贝叶斯神经网络算法和随机森林算法建立光谱信息和土壤湿度数据之间的关系。结果表明:采用贝叶斯线性回归反演时,当隐含层节点个数为24时训练效果最好,R^2为0.755,均方根误差RMSE为0.161;采用随机森林机器学习算法反演时,当决策树个数为60时效果最好,R^2为0.809,均方根误差RMSE为0.120。对"天宫二号"影像进行土壤湿度反演时,随机森林模型比贝叶斯神经网络模型的精度更高,拟合效果更好,可以实现较为准确的大范围土壤水分含量预测。
The model between the spectral information of remote sensing image and the soil moisture is established by means of the neural network and machine learning method to predict the surface soil moisture on a large scale by remote sensing.The visible light and near-infrared spectrum image acquired by the Wide-band imager on September 24,2016 is used as the model input,SMAP/Sentinel-1 L2 soil moisture product with the same time and latitude and longitude as the Tiangong-2 image is selected as the model output,and the relationship between spectral information and soil moisture data is established by means of the Bayesian neural network algorithm and random forest algorithm,respectively.The results show that when using Bayesian linear regression inversion,the training effect is best with the number of hidden layer nodes is 24,R-square is 0.755,and root mean square error is 0.161.In the soil moisture inversion of tiangong-2 image,the random forest model has higher accuracy and better fitting effect than the Bayesian neural network model,which can achieve more accurate prediction of soil moisture content in a large range.
作者
常江
丁雷
CHANG Jiang;DING Lei(CAS Key Laboratory of Infrared Detection and Imaging Technology,CAS Shanghai Institute of Technical Physics,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China;Shanghai University of Science and Technology,Shanghai 200031,China)
出处
《现代电子技术》
北大核心
2020年第6期82-85,89,共5页
Modern Electronics Technique
基金
国家重点研发计划项目(2018YFB0504701)。
