Broadacre Crop Yield Estimation Using Imaging Spectroscopy from Unmanned Aerial Systems (UAS): A Field-Based Case Study with Snap Bean

Accurate, precise, and timely estimation of crop yield is key to a grower’s ability proactively manage growth predict harvest logistics. Such predictions typically are based on multi-parametric models in-situ sampling. Here we investigate the extension greenhouse study, low-altitude unmanned aerial systems (UAS). Our principal objective was snap bean (Phaseolus vulgaris) using imaging spectroscopy (hyperspectral imaging) in visible near-infrared (VNIR; 400–1000 nm) region via UAS. We aimed solve problem modelling by identifying spectral features explaining evaluating best time period for accurate prediction, early time. introduced Python library, named Jostar, feature selection. Embedded proposed new ranking method selected that reaches an agreement between multiple optimization models. Moreover, implemented well-known denoising algorithm data used this study. This study benefited from two years remotely sensed data, captured at instances over summers 2019 2020, with 24 plots 18 plots, respectively. Two stage models, late harvest, were assessed different locations upstate New York, USA. Six varieties quantified components yield, pod weight seed length. vegetation detection algorithms. Red-Edge Normalized Difference Vegetation Index (RENDVI) Spectral Angle Mapper (SAM), subset fields into vs. non-vegetation pixels. Partial least squares regression (PLSR) as model. Among nine embedded Genetic Algorithm (GA), Ant Colony Optimization (ACO), Simulated Annealing (SA), Particle Swarm (PSO) their resulting joint ranking. The findings show can be explained high coefficient determination (R2 = 0.78–0.93) low root-mean-square error (RMSE 940–1369 kg/ha) data. Seed length assessment resulted higher accuracies 0.83–0.98) lower errors 4.245–6.018 mm). used, ACO SA outperformed others SAM approach showed improved results when compared RENDVI dense canopies being examined. Wavelengths 450, 500, 520, 650, 700, 760 nm, identified almost all sets used. 44–55 days after planting (DAP) optimal assessment. Future work should involve transferring learned concepts multispectral system, eventual operational use; further attention also paid ground truth collection technique, since indicator far more rapid straightforward.