ASABE Technical Library - Abstract
Member and Access Notice
Tree-Specific Optimization of Shaking Parameters for Mechanical Harvesting Using a YOLO-Based Data-Driven Model
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: 2026 ASABE Annual International Meeting 2601148.(doi:10.13031/aim.202601148)Authors: Mohsen Farajijalal, Samira Malek, Mohammadmehdi Maharlooei, Joshua H. Viers, Reza Ehsani
Keywords: mechanical harvesting, nut tree, sensor system, trunk shaker, vibration acceleration, wavelets.
Abstract. Mechanical harvesting of nut trees relies heavily on trunk shaker harvesting machines. Their shaking parameters are typically selected based on operator experience, often resulting in suboptimal fruit removal efficiency and potential tree damage. This study presents a data-driven framework for optimizing mechanical harvesting parameters in pistachio orchards by integrating multi-sensor measurements, aerial imaging, and machine learning.
A multi-sensor system, including inertial measurement units, accelerometers, GPS, and displacement sensors, was installed on both the tree and the shaker head to capture vibration responses during harvesting operations. Physical characteristics of trees, such as trunk diameter, height, and canopy area, were quantified using ground measurements and aerial imagery.
A fully connected neural network (FCNN), implemented as a feedforward multilayer perceptron (MLP), was developed to predict optimal shaking parameters, specifically trunk vibration acceleration and shaking duration, based on tree-specific features. In this architecture, each neuron in a given layer is connected to all neurons in the subsequent layer, enabling the model to capture nonlinear relationships between input tree characteristics and output shaking responses. The model was trained and validated using extensive field data collected from pistachio orchards, demonstrating agreement between measured and predicted responses as well as robust generalization to unseen data. In addition, frequency-domain and wavelet analyses were conducted to characterize vibration behavior and provide further insight into energy propagation within the tree structure.
The proposed framework enables adaptive, tree-specific adjustment of shaker parameters, reducing reliance on operator intuition while improving harvesting efficiency and minimizing structural damage to trees. The results demonstrate the potential of data-driven control strategies to advance precision and automation in mechanical orchard harvesting.
(Download PDF) (Export to EndNotes)