Improvement of Unmanned Aerial Vehicle Object Detection Through Scale Optimization and YOLOv7-Tiny Anchor Adjustment

Febrianto Eko Saputra, Ingrid Nurtanio

2025

Abstract

Object detection in Unmanned Aerial Vehicle (UAV) images is challenging due to variations in scale, shooting angle, and object density, particularly for small objects. YOLOv7-Tiny, a lightweight real-time model, offers high efficiency but limited accuracy in this scenario. This study proposes architectural modifications and anchor optimization to enhance detection performance. The architecture is improved by adding a high-resolution detection path to the neck and an additional detection layer to the head, thereby strengthening small object feature representation. Furthermore, anchor box optimization using the K-means algorithm with Manhattan Distance produces anchors that are more representative of the UAV dataset’s object size distribution. Experimental results show that the optimized YOLOv7-Tiny achieves stable precision (95) and recall (96), with the F1-Score increasing from 95 to 96 compared to the baseline. The model also improves mAP at low to medium IoU thresholds, raising the average mAP@50–95 from 71.68 to 72.73. However, performance decreases at high IoU thresholds, and inference time slightly increases due to added complexity. Overall, the proposed approach improves UAV small object detection with a trade-off in processing efficiency.

Paper Citation

in Harvard Style

Eko Saputra F. and Nurtanio I. (2025). Improvement of Unmanned Aerial Vehicle Object Detection Through Scale Optimization and YOLOv7-Tiny Anchor Adjustment. In Proceedings of the 1st International Conference on Research and Innovations in Information and Engineering Technology - Volume 1: RITECH; ISBN 978-989-758-784-9, SciTePress, pages 37-44. DOI: 10.5220/0014268500004928

in Bibtex Style

@conference{ritech25,
author={Febrianto Eko Saputra and Ingrid Nurtanio},
title={Improvement of Unmanned Aerial Vehicle Object Detection Through Scale Optimization and YOLOv7-Tiny Anchor Adjustment},
booktitle={Proceedings of the 1st International Conference on Research and Innovations in Information and Engineering Technology - Volume 1: RITECH},
year={2025},
pages={37-44},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0014268500004928},
isbn={978-989-758-784-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Research and Innovations in Information and Engineering Technology - Volume 1: RITECH
TI - Improvement of Unmanned Aerial Vehicle Object Detection Through Scale Optimization and YOLOv7-Tiny Anchor Adjustment
SN - 978-989-758-784-9
AU - Eko Saputra F.
AU - Nurtanio I.
PY - 2025
SP - 37
EP - 44
DO - 10.5220/0014268500004928
PB - SciTePress