Object detection and edge deployment for robot-assisted aircraft walk-around inspection

Introduction Robot‐assisted aircraft walk-around inspection requires recognition of external inspection targets and efficient processing on onboard hardware. Methods We mapped ten visually observable B737‐300 job‐card targets to detection classes and assembled 3,219 images with 6,808 annotations. Seven YOLO configurations were compared on a sequence‐separated dataset with a held‐out test set of 457 images. Deployment evaluation combined detection accuracy, OpenVINO conversion consistency, and repeated video processing on the robot's onboard computer, an Intel Core i5‐1335U MiniPC. Six of the seven configurations each processed the same 7,255‐frame, 241.82‐s walk‐around recording three times using OpenVINO CPU inference. Results YOLO11s‐P2 achieved the highest mAP50 (0.8939), while YOLO26s achieved the highest mAP50‐95 (0.4308). YOLO11n completed the recording in 220.7 ± 6.5 s (mean ± SD) and was the only configuration to process every frame within the recording duration in all three runs. Discussion These results identify OpenVINO YOLO11n as a practical deployment option for the evaluated onboard workload.

Authors

Institutions

Publication Details

Journal
Frontiers in Mechanical Engineering
Published
2026-11-17
DOI
https://doi.org/10.3389/fmech.2026.2007121
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Object detection and edge deployment for robot-assisted aircraft walk-around inspection

Cao KeLu, Yuying Dong, Weihan Li, Yiming Yang
2 links
Frontiers in Mechanical Engineering
Advanced Neural Network Applications
article

Object detection and edge deployment for robot-assisted aircraft walk-around inspection

Cao KeLu, Yuying Dong, Weihan Li, Yiming Yang
article en

Abstract

Introduction Robot‐assisted aircraft walk-around inspection requires recognition of external inspection targets and efficient processing on onboard hardware. Methods We mapped ten visually observable B737‐300 job‐card targets to detection classes and assembled 3,219 images with 6,808 annotations. Seven YOLO configurations were compared on a sequence‐separated dataset with a held‐out test set of 457 images. Deployment evaluation combined detection accuracy, OpenVINO conversion consistency, and repeated video processing on the robot's onboard computer, an Intel Core i5‐1335U MiniPC. Six of the seven configurations each processed the same 7,255‐frame, 241.82‐s walk‐around recording three times using OpenVINO CPU inference. Results YOLO11s‐P2 achieved the highest mAP50 (0.8939), while YOLO26s achieved the highest mAP50‐95 (0.4308). YOLO11n completed the recording in 220.7 ± 6.5 s (mean ± SD) and was the only configuration to process every frame within the recording duration in all three runs. Discussion These results identify OpenVINO YOLO11n as a practical deployment option for the evaluated onboard workload.

Frontiers in Mechanical EngineeringVol. 12
Xi'an Aeronautical Polytechnic Institute (CN)
Advanced Neural Network Applications
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