A Minimalist Approach to Yield Mapping of Standing Wheat Crop with Unmanned Aerial Vehicles

Yield estimation and mapping of standing crops are often based on tedious data gathering procedures that can be daunting and not cost-effective in the absence of harvester-mounted yield mappers. A cost-effective solution with reasonable accuracy has greater potential for adoption especially if one can leverage latest machine learning tools to supplant tedious processes. This study conducts a feasibility test in using drones in a minimalist sampling strategy to estimate wheat yields over different productivity zones. The first step is to use unsupervised clustering of spatio-temporal multivariate data to delineate zones of homogeneous vegetation vigour. These zones are assumed to capture the variability in yield and aid in designing an efficient sampling strategy. The second step involves using a UAV-mounted camera to capture digital images to estimate the wheat head count and then to derive a yield estimate within the image field of view. Using physical counting of grains within a 0.5 × 0.5 m reference frame, the performance of the estimation procedure was observed. The results show that while the workflow is tractable and friendly in low-resource environments, the accuracy is poor at this stage. Pertinent challenges and potential improvement strategies are discussed.