Answers are grounded in this paper and cite their sources.
Accurate yield mapping is critical for orchard management and understanding within-orchard variability, especially in high-value nut crops such as almonds and pistachios. Although remote sensing and on-harvester solutions have progressed for row crops, tree-level yield mapping for mechanically harvested crops remains unavailable. This study proposes a method for tree-level yield resolution in shake-catch systems where nuts fall onto a conveyor, forming a continuous flow. The main challenge is mixing: low conveyor-belt speed and high harvester speed between trees cause nuts from adjacent trees to overlap on the belt. Although sensing, calibration, and field validation are conducted in almond orchards, the method is formulated to be generalisable to other shake-catch nut crops. Nut volume flow is estimated with a laser and belt-speed encoder. Single-tree yields are obtained by (a) recording flows from individually harvested trees to select a flow “kernel,” and (b) deconvolving continuous flows into positioned, parameterised kernels corresponding to per-tree yields. Kernels tested were Gaussian, Polynomially Modified Gaussian (PMG2), and PMG2 with a Tukey window (PMG2-Tukey) to account for discharge duration. Volume-to-weight calibration using almonds in a controlled environment and three orchards yielded mean absolute percentage errors (MAPE) of 3.37%, 5.81%, 7.50%, and 13.10%. Models are fitted via nonlinear least squares with resampling. For ground-truth (GT) trees, relative error (RE) changed with resampling frequency; PMG2 and PMG2-Tukey outperformed Gaussian, and weighted absolute percentage error (WAPE) was lowest at 1 Hz. Non-GT half-rows showed the smallest WAPE at 1 Hz; for non-GT trees, PMG2 achieved RE as low as 0.39%.
No takes yet. Share an insight, caveat, or question.
Villacrés et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: