A computer vision framework for real-time seed spacing evaluation in seed metering devices using YOLOv12 and SORT

This study presents an automated computer vision framework for seed spacing evaluation on a sticky-belt test stand, integrating YOLOv12-based object detection with the Simple Online and Realtime Tracking (SORT) algorithm. A dataset of 741 manually annotated frames, extracted at 10 FPS from 13 video recordings, was expanded to 1,722 images by augmenting the training subset, and YOLOv8s, YOLOv11s and YOLOv12 (n, s, m, l, x) variants were trained and evaluated. Trials were conducted at four operating speed levels and three target spacings (12.77, 14.10 and 18.25 cm). All detectors reached mAP@50 ≥ 0.994 on held-out frames, and the differences between models (≤ 0.0003) were too small to discriminate among them; YOLOv12m (precision 0.9918, recall 0.9875, mAP@50–95 0.8376) was therefore selected on the basis of an accuracy–efficiency trade-off. The detection and tracking stages required 11.1–11.3 ms per frame on a laptop RTX 4070 GPU (≈89 FPS computational throughput), so that the deployed system is limited by the 30 FPS camera rather than by computation. Planter performance was quantified from 1,186 vision-derived spacings using ISO 7256/1 indices. The quality of feed index decreased from 97.84 % at the lowest to 80.00 % at the highest speed level, while the miss index rose from 1.53 % to 18.08 % and the coefficient of variation from 16.35 % to 39.13 %. A binomial generalized linear model showed that operating speed was the dominant factor affecting miss probability (likelihood-ratio χ 2 (3) = 58.9, p < 0.001; odds ratio 4.60 per 10 Hz increase in inverter frequency), whereas the main effect of target spacing (p = 0.32) and the speed × spacing interaction (p = 0.31) were not statistically significant. Under the tested laboratory configuration, the YOLOv12–SORT framework provides an objective and automated alternative to manual sticky-belt spacing measurement; field validation and seed-by-seed comparison with an independent reference remain necessary before broader use.

Authors

Institutions

Publication Details

Journal
Engineering Science and Technology an International Journal
Published
2026-11-19
DOI
https://doi.org/10.1016/j.jestch.2026.102591
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A computer vision framework for real-time seed spacing evaluation in seed metering devices using YOLOv12 and SORT

Keziban Yalçın Dokumacı
1 links
Engineering Science and Technology an International Journal
Smart Agriculture and AI
article

A computer vision framework for real-time seed spacing evaluation in seed metering devices using YOLOv12 and SORT

Keziban Yalçın Dokumacı
article en

Abstract

This study presents an automated computer vision framework for seed spacing evaluation on a sticky-belt test stand, integrating YOLOv12-based object detection with the Simple Online and Realtime Tracking (SORT) algorithm. A dataset of 741 manually annotated frames, extracted at 10 FPS from 13 video recordings, was expanded to 1,722 images by augmenting the training subset, and YOLOv8s, YOLOv11s and YOLOv12 (n, s, m, l, x) variants were trained and evaluated. Trials were conducted at four operating speed levels and three target spacings (12.77, 14.10 and 18.25 cm). All detectors reached mAP@50 ≥ 0.994 on held-out frames, and the differences between models (≤ 0.0003) were too small to discriminate among them; YOLOv12m (precision 0.9918, recall 0.9875, mAP@50–95 0.8376) was therefore selected on the basis of an accuracy–efficiency trade-off. The detection and tracking stages required 11.1–11.3 ms per frame on a laptop RTX 4070 GPU (≈89 FPS computational throughput), so that the deployed system is limited by the 30 FPS camera rather than by computation. Planter performance was quantified from 1,186 vision-derived spacings using ISO 7256/1 indices. The quality of feed index decreased from 97.84 % at the lowest to 80.00 % at the highest speed level, while the miss index rose from 1.53 % to 18.08 % and the coefficient of variation from 16.35 % to 39.13 %. A binomial generalized linear model showed that operating speed was the dominant factor affecting miss probability (likelihood-ratio χ 2 (3) = 58.9, p < 0.001; odds ratio 4.60 per 10 Hz increase in inverter frequency), whereas the main effect of target spacing (p = 0.32) and the speed × spacing interaction (p = 0.31) were not statistically significant. Under the tested laboratory configuration, the YOLOv12–SORT framework provides an objective and automated alternative to manual sticky-belt spacing measurement; field validation and seed-by-seed comparison with an independent reference remain necessary before broader use.

Engineering Science and Technology an International JournalVol. 84
Selçuk University (TR)
Selçuk Üniversitesi
Zero hunger
Smart Agriculture and AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon