基于遗传优化神经网络的多源遥感数据反演土壤水分

为快速反演较高精度土壤水分,提出用遗传算法优化后的神经网络辅以多源遥感数据的方法进行地表土壤水分反演。首先建立4层神经网络并用遗传算法优化此网络,之后以雷达数据不同极化(VV、VH、VH/VV)的后向散射系数、雷达入射角、光学数据的归一化植被指数(NDVI)、以及高程数据作为网络的输入,土壤水分数据为输出,对网络进行训练与仿真,再运用地表实际测量数据与反演数据做对比验证。结果表明:反演结果与实际测量数据相关性良好,R2可达0. 79。采用遗传算法对神经网络优化的土壤水分反演方法可行,且添加光学数据等辅助数据后土壤水分反演效果更优,为多源遥感土壤水分的协同反演研究提供新思路。

Soil moisture inversion based on genetic optimization neural network and multi-source remote sensing data

In order to quickly invert higher precision soil moisture, this paper used the genetic algorithm optimized neural network and multi-source remote sensing data to invert the surface soil moisture. First, a four-layer neural network was established and the network was optimized by genetic algorithm. Then, we used the backscattering coefficient of radar data with different polarizations (VV, VH, VH/VV), radar incident angle, normalized vegetation index (NDVI) of optical data, and elevation data as input to neural network, the soil moisture data as the output to train and simulate the network. Finally, the inversion data was verified by the actual measured data of the surface. The results showed that the correlation between the inversion data and the measured data is high (R²=0.79). The method of genetic algorithm optimized neural network is feasible to calculate soil moisture content, and the soil water inversion effect is better after adding auxiliary data such as optical data. This study provides a new idea for the collaborative inversion of soil moisture in the multi-source remote sensing.