This paper presents a high-throughput method for Above Ground Estimation of Biomass (AGBE) in rice using multispectral near-infrared (NIR) imagery captured at different scales of the crop. By developing an integrated aerial crop monitoring solution using an Unmanned Aerial Vehicle (UAV), our approach calculates 7 vegetation indices that are combined in the form of multivariable regressions depending on the stage of rice growth: vegetative, reproductive or ripening. We model the relationship of these vegetation indices to estimate the biomass of a certain crop area. The methods are calibrated by using a minimum sampling area of 1 linear meter of the crop. Comprehensive experimental tests have been carried out over two different rice varieties under upland and lowland rice production systems. Results show that the proposed approach is able to estimate the biomass of large areas of the crop with an average correlation of 0.76 compared with the traditional manual destructive method. To our knowledge, this is the first work that uses a small sampling area of 1 linear meter to calibrate and validate NIR image-based estimations of biomass in rice crops.

Leaf nitrogen (N) directly correlates to chlorophyll production, affecting crop growth and yield. Farmers use soil plant analysis development (SPAD) devices to calculate the amount of chlorophyll present in plants. However, monitoring large-scale crops using SPAD is prohibitively time-consuming and demanding. This paper presents an unmanned aerial vehicle (UAV) solution for estimating leaf N content in rice crops, from multispectral imagery. Our contribution is twofold: (i) a novel trajectory control strategy to reduce the angular wind-induced perturbations that affect image sampling accuracy during UAV flight, and (ii) machine learning models to estimate the canopy N via vegetation indices (VIs) obtained from the aerial imagery. This approach integrates an image processing algorithm using the GrabCut segmentation method with a guided filtering refinement process, to calculate the VIs according to the plots of interest. Three machine learning methods based on multivariable linear regressions (MLR), support vector machines (SVM), and neural networks (NN), were applied and compared through the entire phonological cycle of the crop: vegetative (V), reproductive (R), and ripening (Ri). Correlations were obtained by comparing our methods against an assembled ground-truth of SPAD measurements. The higher N correlations were achieved with NN: $0.98$ (V), $0.94$ (R), and $0.89$ (Ri). We claim that the proposed UAV stabilization control algorithm significantly improves on the N-to-SPAD correlations by minimizing wind perturbations in real-time and reducing the need for offline image corrections.

The use of Unmanned Aerial Vehicle (UAV) images for biomass and nitrogen estimation offers multiple opportunities for improving rice yields. UAV images provide detailed, high-resolution visual information about vegetation properties, enabling the identification of phenotypic characteristics for selecting the best varieties, improving yield predictions, and supporting ecosystem monitoring and conservation efforts. In this study, an analysis of biomass and nitrogen is conducted on 59 rice plots selected at random from a more extensive trial comprising 400 rice genotypes. A UAV acquires multispectral reflectance channels across a rice field of subplots containing different genotypes. Based on the ground-truth data, yields are characterized for the 59 plots and correlated with the Vegetation Indices (VIs) calculated from the photogrammetric mapping. The VIs are weighted by the segmentation of the plants from the soil and used as a feature matrix to estimate, via machine learning models, the biomass and nitrogen of the selected rice genotypes. The genotype IR 93346 presented the highest yield with a biomass gain of 10,252.78 kg/ha and an average daily biomass gain above 49.92 g/day. The VIs with the highest correlations with the ground-truth variables were NDVI and SAVI for wet biomass, GNDVI and NDVI for dry biomass, GNDVI and SAVI for height, and NDVI and ARVI for nitrogen. The machine learning model that performed best in estimating the variables of the 59 plots was the Gaussian Process Regression (GPR) model with a correlation factor of 0.98 for wet biomass, 0.99 for dry biomass, and 1 for nitrogen. The results presented demonstrate that it is possible to characterize the yields of rice plots containing different genotypes through ground-truth data and VIs.

In Colombia up to 40% of yield variability is due to the effects of climate variations. Rapid phenotyping methods are needed to properly assess the crop and improve production rates. In this paper, we propose to focus on developing a noninvasive system for speeding up monitoring tasks in rice crops. Unmanned Aerial Vehicles are used to gather multispectral visual information for high-throughput crop monitoring. Geo-referenced digital surface models of the crop are generated based on image mosaicing techniques to allow for the autonomous computation of several vegetative indices. This paper presents the implemented system (hardware and software) and a field report of experiments carried out at different crop growth stages.