Estimating LAI From Winter Wheat Using UAV Data and CNNs

With the advent of high-resolution unmanned aerial vehicle (UAV) data and advancing methods of deep learning, new opportunities have emerged in remote sensing to assess biophysical plant parameters. In this study, we investigated the potential of UAV-borne RGB data and convolutional neural networks (CNNs) to estimate the leaf area index (LAI) of winter wheat during two cropping seasons. In this context, spectral RGB and geometric plant information based on a normalized surface model (nDSM) were used as input variables. The results of the study demonstrated the suitability of optical UAV data and CNNs for LAI estimation of winter wheat at different growth stages and under various lightning conditions. The combination of RGB data and plant structures provided the best overall prediction accuracy ( $r^{2} = 0.83$ ) compared to the models with only one input source (RGB: $r^{2} = 0.58$ , nDSM: $r^{2} = 0.75$ ). Especially the estimation of low and high LAI values was improved using the complementary image information. Moreover, the results showed that the CNN models outperformed two classical machine learning (ML) approaches in terms of accuracy.