Deep-Learning-Based Multispectral Image Reconstruction from Single Natural Color RGB Image—Enhancing UAV-Based Phenotyping

Multispectral images (MSIs) are valuable for precision agriculture due to the extra spectral information acquired compared natural color RGB (ncRGB) images. In this paper, we thus aim generate high spatial MSIs through a robust, deep-learning-based reconstruction method using ncRGB Using data from agronomic research trial maize and breeding rice, first reproduced rendering model, Model-True image (Model-TN), which was built benchmark hyperspectral dataset. Subsequently, an MSI Model-Natural (Model-NM), trained based on prepared (ncRGB-Con) pairs, ensuring model can use widely available as input. The integrated loss function of mean relative absolute error (MRAEloss) divergence (SIDloss) were most effective during building both models, while models MRAEloss more robust towards variability between growing seasons species. reliability reconstructed demonstrated by coefficients determination ground truth values, Normalized Difference Vegetation Index (NDVI) example. advantages “reconstructed” NDVI over Triangular Greenness (TGI), calculated directly images, illustrated their higher capabilities in differentiating three levels irrigation treatments plants. This study emphasizes that performance could benefit optimized intermediate step preparation. ability developed reconstruct high-quality low-cost will, particular, promote application plant phenotyping agriculture.