Soil moisture content estimation in winter wheat planting area for multi-source sensing data using CNNR

Rapid and accurate estimation of soil moisture content (SMC) is an important part of precision agriculture, and it is also one of the key problems to be solved in field real-time monitoring and precision irrigation. Most of the existing studies are limited to SMC monitoring of bare soil, which can be obtained by optical remote sensing (e.g., multispectral, hyperspectral) or Synthetic Aperture Radar (SAR). However, for the soil covered by vegetation, such as farmland, there are some theoretical defects with only one of the measuring methods. Meanwhile, in order to break through the limitations of low spatial and temporal resolutions of satellite remote sensing, it is of great significance to study SMC retrieval based on multi-source remote sensing data for the near earth UAV remote sensing systems. Based on this, this paper, taking the winter wheat planting area in Guanzhong plain of China as the research area, combines the advantages of ultra-wideband (UWB) radar, and multispectral remote sensing data, to reduce the influences of vegetation coverage on the estimation accuracy. A one-dimensional regression convolution neural network model is constructed to realize the quantitative prediction and estimation of SMC in farmland. The carried out experiments show that the proposed CNNR model has a better performance than traditional SVR and GRNN models and the R-2, RMSE and RPD are 0.7453, 0.0140 cm(3)/cm(3) and 1.9246, respectively. After introducing NDVI, MSAVI and DVI vegetation indices generated from multispectral images, the accuracy of the three models increased significantly. Among the three models, the constructed CNNR model has the best performance, and its R-2, RMSE and RPD reach 0.9168, 0.0089 cm(3)/cm(3), and 3.0201. Furthermore, after adding different levels of Gaussian noise to the original radar echoes, the CNNR model constructed in this paper still has the highest prediction accuracy and the strongest noise robustness.