Work

Improved Estimation of Soil-to-Sugar Beet Ratio Using Deep Learning and Domain Adaptation for Precision Agriculture

Year: 2026

Type: conference-paper

Abstract: Accurate estimation of soil adhering to sugar beets is important for both sugar factories and farmers.Since weighing each truckload before and after cleaning is impractical, soil tare is typically estimated visually by trained evaluators.However, Gage R&R and accuracy analyses revealed high variability in these manual assessments, highlighting the need for a more reliable and objective solution.This study proposes an automated pipeline for soil-to-beet ratio estimation while addressing two major challenges: the absence of image-level ground truth field data and the difficulty of manually annotating soil regions.The pipeline combines sugar beet segmentation using Detectron2 with soil segmentation based on intensity and texture analysis.Extracted features are then used to train a Semi-Supervised Domain-Adversarial Neural Network (SS-DANN) to handle domain shifts between laboratory and field datasets.The proposed approach achieved a deviation of 0.39% from ground truth, outperforming human estimators with an average deviation of 1.91%.

Source: Annals of Computer Science and Information Systems

DOI: 10.15439/2026f2778

Authors: Elham Alfuqara, Matthias Hien, Raphael Roeser-Mueller, Johann Brunner, Tobias Dobmeier

Institution: Deggendorf Institute of Technology

Language: English

Topic: Smart Agriculture and AI

Subfield: Plant Science

Field: Agricultural and Biological Sciences

Domain: Life Sciences

SDG: Zero hunger

Open Access status: diamond

OpenAlex ID: W7214399298