BP神经网络辅助的GNSS反射信号NDVI反演

NDVI Inversion of GNSS Reflected Signal Assisted by BP Neural Network

归一化植被指数(normalized difference vegetation index,NDVI)是一种能反映地表植被生长情况和覆盖度的重要指标,针对如何确定研究区域归一化植被指数变化趋势的问题,提出一种BP神经网络辅助的GNSS卫星反射信号NDVI反演方法。从PBO观测网P037和P39站点信噪比观测数据提取的振幅参数作为输入值,归一化植被指数作为输出值,构建BP神经网络辅助的GNSS卫星反射信号植被指数反演模型,并与线性回归模型进行对比,实验结果显示:P037和P039站点振幅线性回归的相关系数为0.7003和0.7756,均方根误差为0.0622和0.0760,BP模型的相关系数为0.8023和0.8394,均方根误差为0.0336和0.0459,表明BP神经网络辅助的GNSS卫星反射信号反演模型获取的归一化植被指数优于线性回归模型,为获取准实时、低成本和高时间分辨率的NDVI提供了新的思路,证明了该方法的可行性。

The normalized difference vegetation index(NDVI)is an important indicator that reflects the growth and coverage of surface vegetation.Aiming at the problem of determining the trend of normalized vegetation index in the study area,a BP neural network-assisted NDVI inversion method for GNSS satellite reflection signal is proposed.The amplitude parameters extracted from the signal-to-noise ratio observation data of P037 and P39 stations of PBO observation network are used as input values,and the vegetation index is used as the output value to construct the BP neural network-assisted GNSS satellite reflection signal vegetation index inversion model and comparing it with linear regression model.The result shows that the correlation coefficients of amplitude linear regression of P037 and P039 stations are 0.7003 and 0.7756,and the root mean square errors are 0.0622 and 0.0760.The correlation coefficients of the BP model are 0.8023 and 0.8394,and the root mean square error is 0.0336 and 0.0459.It is shown that the normalized vegetation index obtained by the BP neural network-assisted GNSS satellite reflection signal inversion model is better than the linear regression model,it provides a new idea to obtain NDVI with real-time,low cost and high time resolution,which proves the feasibility of this method.