Orchard digital twin: A prototype for smart agricultural monitoring

This study presents a method for creating detailed digital twins (DT) of apple orchards using an unmanned ground vehicle (UGV) and a combination of sensors, including a handheld laser scanner (HLS), a low-cost depth camera, and a Global Navigation Satellite Systems (GNSS) receiver. We aim to provide an updatable 3D geospatial database to automate agricultural processes, such as apple monitoring and collection. The workflow was tested in a portion of an apple orchard in northwest Italy and is comprised of four phases. Phase 1 involves sensor-integration on the vehicle and data acquisition. Phase 2 defines and stores the orchard's geometries in a georeferenced 3D model; this phase also includes the segmentation of individual trees within each row. Phase 3 detects and segments apples using an artificial intelligence (AI) algorithm applied to RGB images captured by the depth camera; segmented apples are then projected onto the 3D model. Our work demonstrates how unmanned ground vehicles (UGVs) integrated with sensors can be applied to create a detailed, updatable orchard DT, a tool which can inform and automate agricultural tasks, ultimately increasing efficiency and reducing waste.